ICAART 2022 Abstracts


Area 1 - Artificial Intelligence

Full Papers
Paper Nr: 6
Title:

Time Series Augmentation based on Beta-VAE to Improve Classification Performance

Authors:

Domen Kavran, Borut Žalik and Niko Lukač

Abstract: Classification models that provide good generalization are trained with sufficiently large datasets, but these are often not available due to restrictions and limited resources. A novel augmentation method is presented for generating synthetic time series with Beta-VAE variational autoencoder, which has ResNet-18 inspired architecture. The proposed augmentation method was tested on benchmark univariate time series datasets. For each dataset, multiple variational autoencoders were used to generate different amounts of synthetic time series samples. These were then used, along with the original train set samples, to train MiniRocket classification models. By using the proposed augmentation method, a maximum increase of 1,22% in classification accuracy was achieved on the tested datasets in comparison to baseline results, which were obtained by training only with original train sets. An increase of up to 0,81% in accuracy of simple machine learning classifiers was observed by benchmarking the proposed augmentation method with the 1-nearest neighbor algorithm.
Download

Paper Nr: 8
Title:

Leveraging Causal Relations to Provide Counterfactual Explanations and Feasible Recommendations to End Users

Authors:

Riccardo Crupi, Beatriz San Miguel González, Alessandro Castelnovo and Daniele Regoli

Abstract: Over the last years, there has been a growing debate on the ethical issues of Artificial Intelligence (AI). Explainable Artificial Intelligence (XAI) has appeared as a key element to enhance trust of AI systems from both technological and human-understandable perspectives. In this sense, counterfactual explanations are becoming a de facto solution for end users to assist them in acting to achieve a desired outcome. In this paper, we present a new method called Counterfactual Explanations as Interventions in Latent Space (CEILS) to generate explanations focused on the production of feasible user actions. The main features of CEILS are: it takes into account the underlying causal relations by design, and can be set on top of an arbitrary counterfactual explanation generator. We demonstrate how CEILS succeeds through its evaluation on a real dataset of the financial domain.
Download

Paper Nr: 11
Title:

Bilinear Multi-Head Attention Graph Neural Network for Traffic Prediction

Authors:

Haibing Hu, Kai Han and Zhizhuo Yin

Abstract: Traffic forecasting is an important component of Intelligent Transportation System (ITS) and it has the significance for reducing traffic accidents and improving public safety. Due to the complex spatial-temporal dependencies and the uncertainty of road network, the research on this problem is quite challenging. Some of the latest studies utilize graph convolutional networks (GCNs) to model spatial-temporal relationships. However, these methods are only based on the linear weighted summation of the neighborhood to form the representation of the target node, which cannot capture the signal between pairwise node interactions. In many scenes, adding pairwise node interaction features is an essential way to better represent the target node. Therefore, in this article, we propose an end-to-end novel framework named Bilinear Multi-Head Attention Graph Neural Network (BMHA-GNN) for traffic prediction. We propose a new aggregation operator which utilizes the weighted sum of pairwise interactions of the neighbour nodes and improves the representation ability of GCN based models. We adopt the encoder-decoder framework, the encoder module outputs the representation of traffic data, and the decoder module outputs the prediction results. The multi-head attention mechanism is introduced to aggregate information of different neighbour nodes automatically and stabilize the training process. Extensive experiments are conducted on two real-world datasets (METR-LA, PEMS-BAY) showing that the proposed model BMHA-GNN achieves the state-of-the-art results.
Download

Paper Nr: 12
Title:

Deceptive AI Explanations: Creation and Detection

Authors:

Johannes Schneider, Christian Meske and Michalis Vlachos

Abstract: Artificial intelligence (AI) comes with great opportunities but can also pose significant risks. Automatically generated explanations for decisions can increase transparency and foster trust, especially for systems based on automated predictions by AI models. However, given, e.g., economic incentives to create dishonest AI, to what extent can we trust explanations? To address this issue, our work investigates how AI models (i.e., deep learning, and existing instruments to increase transparency regarding AI decisions) can be used to create and detect deceptive explanations. As an empirical evaluation, we focus on text classification and alter the explanations generated by GradCAM, a well-established explanation technique in neural networks. Then, we evaluate the effect of deceptive explanations on users in an experiment with 200 participants. Our findings confirm that deceptive explanations can indeed fool humans. However, one can deploy machine learning (ML) methods to detect seemingly minor deception attempts with accuracy exceeding 80% given sufficient domain knowledge. Without domain knowledge, one can still infer inconsistencies in the explanations in an unsupervised manner, given basic knowledge of the predictive model under scrutiny.
Download

Paper Nr: 16
Title:

LABNET: A Collaborative Method for DNN Training and Label Aggregation

Authors:

Amirmasoud Ghiassi, Robert Birke and Lydia Y. Chen

Abstract: Today, to label the massive datasets needed to train Deep Neural Networks (DNNs), cheap and error-prone methods such as crowdsourcing are used. Label aggregation methods aim to infer the true labels from noisy labels annotated by crowdsourcing workers via labels statistics features. Aggregated labels are the main data source to train deep neural networks, and their accuracy directly affects the deep neural network performance. In this paper, we argue that training DNN and aggregating labels are not two separate tasks. Incorporation between DNN training and label aggregation connects data features, noisy labels, and aggregated labels. Since each image contains valuable knowledge about its label, the data features help aggregation methods enhance their performance. We propose LABNET an iterative two-step method. Step one: the label aggregation algorithm provides labels to train the DNN. Step two: the DNN shares a representation of the data features with the label aggregation algorithm. These steps are repeated until the converging label aggregation error rate. To evaluate LABNET we conduct an extensive empirical comparison on CIFAR-10 and CIFAR-100 under different noise and worker statistics. Our evaluation results show that LABNET achieves the highest mean accuracy with an increase of at least 8% to 0.6% and lowest error rate with a reduction of 7.5% to 0.25% against existing aggregation and training methods in most cases.
Download

Paper Nr: 28
Title:

RVPLAN: Runtime Verification of Assumptions in Automated Planning

Authors:

Angelo Ferrando and Rafael C. Cardoso

Abstract: Automated planners use a model of the system and apply state transition functions to find a sequence of actions (a plan) that successfully solves a (set of) goal(s). The model used during planning can be imprecise, either due to a mistake at design time or because the environment is dynamic and changed before or during the plan execution. In this paper we use runtime monitors to verify the assumptions of plans at runtime in order to effectively detect plan failures. This paper offers three main contributions: (a) two methods (instantiated and parameterised) to automatically synthesise runtime monitors by translating planning models (STRIPS-like) to temporal logics (Past LTL or Past FO-LTL); (b) an approach to use the resulting runtime monitors to detect failures in the plan; and (c) the RVPLAN tool, which implements (a) and (b). We illustrate the use of our work with a remote inspection running example as well as quantitative results comparing the performance of the proposed monitor generation methods in terms of property synthesis, monitor synthesis, and runtime verification.
Download

Paper Nr: 31
Title:

A Mediator Agent based on Multi-Context System and Information Retrieval

Authors:

Rodrigo Rodrigues, Ricardo A. Silveira and Rafael De Santiago

Abstract: Nowadays, decisions derived from intelligent systems frequently affect human lives (e.g., medicine, robotics, or finance). Traditionally, these systems can be implemented using symbolic or connectionist methods. Since both methods have crucial limitations in different aspects, integrating these methods represents a relevant step to deploying intelligent systems in real-world scenarios. We start tackling the integration of both methods by exploring how to use different types of information during the agent’s decision-making. We modeled and implemented an intelligent agent based on a Multi-Context System (MCS). MCSs allow the representation of information exchange among heterogeneous sources. We use a framework called Sigon to implement the proposed agent. Sigon is a novel framework that enables the development of MCS agents at a programming language level. As a case study, we present a mediator agent for conflict resolution during negotiation. The mediator agent creates advice by retrieving information from the web and employing different data types ( e.g., text and image) during its decision-making. This work provides a promising and flexible way of integrating different information and resources using MCS as the main result.
Download

Paper Nr: 34
Title:

Expert-guided Symmetry Detection in Markov Decision Processes

Authors:

Giorgio Angelotti, Nicolas Drougard and Caroline C. Chanel

Abstract: Learning a Markov Decision Process (MDP) from a fixed batch of trajectories is a non-trivial task whose outcome’s quality depends on both the amount and the diversity of the sampled regions of the state-action space. Yet, many MDPs are endowed with invariant reward and transition functions with respect to some transformations of the current state and action. Being able to detect and exploit these structures could benefit not only the learning of the MDP but also the computation of its subsequent optimal control policy. In this work we propose a paradigm, based on Density Estimation methods, that aims to detect the presence of some already supposed transformations of the state-action space for which the MDP dynamics is invariant. We tested the proposed approach in a discrete toroidal grid environment and in two notorious environments of OpenAI’s Gym Learning Suite. The results demonstrate that the model distributional shift is reduced when the dataset is augmented with the data obtained by using the detected symmetries, allowing for a more thorough and data-efficient learning of the transition functions.
Download

Paper Nr: 35
Title:

Creativity of Deep Learning: Conceptualization and Assessment

Authors:

Marcus Basalla, Johannes Schneider and Jan V. Brocke

Abstract: While the potential of deep learning(DL) for automating simple tasks is already well explored, recent research has started investigating the use of deep learning for creative design, both for complete artifact creation and supporting humans in the creation process. In this paper, we use insights from computational creativity to conceptualize and assess current applications of generative deep learning in creative domains identified in a literature review. We highlight parallels between current systems and different models of human creativity as well as their shortcomings. While deep learning yields results of high value, such as high-quality images, their novelty is typically limited due to multiple reasons such a being tied to a conceptual space defined by training data. Current DL methods also do not allow for changes in the internal problem representation, and they lack the capability to identify connections across highly different domains, both of which are seen as major drivers of human creativity.
Download

Paper Nr: 37
Title:

Unsupervised Learning of State Representation using Balanced View Spatial Deep InfoMax: Evaluation on Atari Games

Authors:

Menore T. Mengistu, Getachew Alemu, Pierre Chevaillier and Pierre De Loor

Abstract: In this paper, we present an unsupervised state representation learning of spatio-temporally evolving sequences of autonomous agents’ observations. Our method uses contrastive learning through mutual information (MI) maximization between a sample and the views derived through selection of pixels from the sample and other randomly selected negative samples. Our method employs balancing MI by finding the optimal ratios of positive-to-negative pixels in these derived (constructed) views. We performed several experiments and determined the optimal ratios of positive-to-negative signals to balance the MI between a given sample and the constructed views. The newly introduced method is named as Balanced View Spatial Deep InfoMax (BVS-DIM). We evaluated our method on Atari games and performed comparisons with the state-of-the-art unsupervised state representation learning baseline method. We show that our solution enables to successfully learn state representations from sparsely sampled or randomly shuffled observations. Our BVS-DIM method also marginally enhances the representation powers of encoders to capture high-level latent factors of the agents’ observations when compared with the baseline method.
Download

Paper Nr: 38
Title:

Multi-agent Transfer Learning in Reinforcement Learning-based Ride-sharing Systems

Authors:

Alberto Castagna and Ivana Dusparic

Abstract: Reinforcement learning (RL) has been used in a range of simulated real-world tasks, e.g., sensor coordination, traffic light control, and on-demand mobility services. However, real world deployments are rare, as RL struggles with dynamic nature of real world environments, requiring time for learning a task and adapting to changes in the environment. Transfer Learning (TL) can help lower these adaptation times. In particular, there is a significant potential of applying TL in multi-agent RL systems, where multiple agents can share knowledge with each other, as well as with new agents that join the system. To obtain the most from inter-agent transfer, transfer roles (i.e., determining which agents act as sources and which as targets), as well as relevant transfer content parameters (e.g., transfer size) should be selected dynamically in each particular situation. As a first step towards fully dynamic transfers, in this paper we investigate the impact of TL transfer parameters with fixed source and target roles. Specifically, we label every agent-environment interaction with agent’s epistemic confidence, and we filter the shared examples using varying threshold levels and sample sizes. We investigate impact of these parameters in two scenarios, a standard predator-prey RL benchmark and a simulation of a ride-sharing system with 200 vehicle agents and 10,000 ride-requests.
Download

Paper Nr: 40
Title:

Semantically Consistent Image-to-Image Translation for Unsupervised Domain Adaptation

Authors:

Stephan Brehm, Sebastian Scherer and Rainer Lienhart

Abstract: Unsupervised Domain Adaptation (UDA) aims to adapt models trained on a source domain to a new target domain where no labelled data is available. In this work, we investigate the problem of UDA from a synthetic computer-generated domain to a similar but real-world domain for learning semantic segmentation. We propose a semantically consistent image-to-image translation method in combination with a consistency regularisation method for UDA. We overcome previous limitations on transferring synthetic images to real looking images. We leverage pseudo-labels in order to learn a generative image-to-image translation model that receives additional feedback from semantic labels on both domains. Our method outperforms state-of-the-art methods that combine image-to-image translation and semi-supervised learning on relevant domain adaptation benchmarks, i.e., on GTA5 to Cityscapes and SYNTHIA to Cityscapes.
Download

Paper Nr: 44
Title:

Safe Policy Improvement Approaches on Discrete Markov Decision Processes

Authors:

Philipp Scholl, Felix Dietrich, Clemens Otte and Steffen Udluft

Abstract: Safe Policy Improvement (SPI) aims at provable guarantees that a learned policy is at least approximately as good as a given baseline policy. Building on SPI with Soft Baseline Bootstrapping (Soft-SPIBB) by Nadjahi et al., we identify theoretical issues in their approach, provide a corrected theory, and derive a new algorithm that is provably safe on finite Markov Decision Processes (MDP). Additionally, we provide a heuristic algorithm that exhibits the best performance among many state of the art SPI algorithms on two different benchmarks. Furthermore, we introduce a taxonomy of SPI algorithms and empirically show an interesting property of two classes of SPI algorithms: while the mean performance of algorithms that incorporate the uncertainty as a penalty on the action-value is higher, actively restricting the set of policies more consistently produces good policies and is, thus, safer.
Download

Paper Nr: 52
Title:

Automated Information Leakage Detection: A New Method Combining Machine Learning and Hypothesis Testing with an Application to Side-channel Detection in Cryptographic Protocols

Authors:

Pritha Gupta, Arunselvan Ramaswamy, Jan P. Drees, Eyke Hüllermeier, Claudia Priesterjahn and Tibor Jager

Abstract: Due to the proliferation of a large amount of publicly available data, information leakage (IL) has become a major problem. IL occurs when secret (sensitive) information of a system is inadvertently disclosed to unauthorized parties through externally observable information. Standard statistical approaches estimate the mutual information between observable (input) and secret information (output), which tends to be a difficult problem for high-dimensional input. Current approaches based on (supervised) machine learning using the accuracy of predictive models on extracted system input and output have proven to be more effective in detecting these leakages. However, these approaches are domain-specific and fail to account for imbalance in the dataset. In this paper, we present a robust autonomous approach to detecting IL, which blends machine learning and statistical techniques, to overcome these shortcomings. We propose to use Fisher’s Exact Test (FET) on the evaluated confusion matrix, which inherently takes the imbalances in the dataset into account. As a use case, we consider the problem of detecting padding side-channels or ILs in systems implementing cryptographic protocols. In an extensive experimental study on detecting ILs in synthetic and real-world scenarios, our approach outperforms the state of the art.
Download

Paper Nr: 60
Title:

Tuniset: The Tunisian Dataset for Emotion Analysis on Online Social Networks

Authors:

Semeh Ben Salem, Sami Naouali and Zied Chtourou

Abstract: Data contained in Online Social Media (OSM) platforms represents a huge source of information that is widely used by several entities to extract valuable information in many domains. These OSM permit sharing ideas, thoughts and news with concise and precise content in text, image or video format. One of the Artificial Intelligence fields developed around OSM is sentiment analysis that identifies the emotion a certain user can develop towards a certain topic. Implementing a sentiment analysis platform to detect the sentiment in a given comment requires developing Machine and Deep Learning algorithms and collecting a huge amount of data to train. In this paper, we propose a special dataset that can be used to identify the emotion such as happiness, anger, neutrality and sadness and train Machine Learning algorithms. This dataset corresponds to a specific language which is the Tunisian dialect. We also provide the whole pipeline used to clean and preprocess the collected data before feeding Machine Learning models.

Paper Nr: 61
Title:

Multi-class Sentiment Classification with Machine Learning on Online Social Networks for the Tunisian Dialect

Authors:

Semeh Ben Salem, Sami Naouali and Zied Chtourou

Abstract: Online Social Networks (OSN) platforms are considered as the most widely used virtual spaces where users share their thoughts and express their sentiments without any restrictions. Text and sentiment analysis gained increasing popularity in the Natural Language Processing (NLP) to deal with various applications. These tasks become more complex when it comes to considering specific language dialects such as Arabic. In this paper, we propose a Tunisian dialect dataset called Tuniset that can be used for emotion classification in order to detect the sentiment within a comment on OSN. Tuniset was experimented using several Machine and Deep Learning models with intensive experiments to see how it performs when detecting sentiments online.

Paper Nr: 65
Title:

Clustering Quality of a High-dimensional Service Monitoring Time-series Dataset

Authors:

Farzana Anowar, Samira Sadaoui and Hardik Dalal

Abstract: Our study evaluates the quality of a high-dimensional time-series dataset gathered from service observability and monitoring application. We construct the target dataset by extracting heterogeneous sub-datasets from many servers, tackling data incompleteness in each sub-dataset using several imputation techniques, and fusing all the optimally imputed sub-datasets. Based on robust data clustering approaches and metrics, we thoroughly assess the quality of the initial dataset and the reconstructed datasets produced with Deep and Convolutional AutoEncoders. The experiments reveal that the Deep AutoEncoder dataset’s performances outperform the initial dataset’s performances.
Download

Paper Nr: 76
Title:

DLVGen: A Dual Latent Variable Approach to Personalized Dialogue Generation

Authors:

Jing Yang Lee, Kong Aik Lee and Woon Seng Gan

Abstract: The generation of personalized dialogue is vital to natural and human-like conversation. Typically, personalized dialogue generation models involve conditioning the generated response on the dialogue history and a representation of the persona/personality of the interlocutor. As it is impractical to obtain the persona/personality representations for every interlocutor, recent works have explored the possibility of generating personalized dialogue by finetuning the model with dialogue examples corresponding to a given persona instead. However, in real-world implementations, a sufficient number of corresponding dialogue examples are also rarely available. Hence, in this paper, we propose a Dual Latent Variable Generator (DLVGen) capable of generating personalized dialogue in the absence of any persona/personality information or any corresponding dialogue examples. Unlike prior work, DLVGen models the latent distribution over potential responses as well as the latent distribution over the agent’s potential persona. During inference, latent variables are sampled from both distributions and fed into the decoder. Empirical results show that DLVGen is capable of generating diverse responses which accurately incorporate the agent’s persona.
Download

Paper Nr: 78
Title:

Learning Heuristic Estimates for Planning in Grid Domains by Cellular Simultaneous Recurrent Networks

Authors:

Michaela Urbanovská and Antonín Komenda

Abstract: Automated planning provides a powerful general problem solving tool, however, its need for a model creates a bottleneck that can be an obstacle for using it in real-world settings. In this work we propose to use neural networks, namely Cellular Simultaneous Recurrent Networks (CSRN), to process a planning problem and provide a heuristic value estimate that can more efficiently steer the automated planning algorithms to a solution. Using this particular architecture provides us with a scale-free solution that can be used on any problem domain represented by a planar grid. We train the CSRN architecture on two benchmark domains, provide analysis of its generalizing and scaling abilities. We also integrate the trained network into a planner and compare its performance against commonly used heuristic functions.
Download

Paper Nr: 79
Title:

Method for Improving Quality of Adversarial Examples

Authors:

Duc-Anh Nguyen, Kha Do Minh, Duc-Anh Pham and Pham N. Hung

Abstract: To evaluate the robustness of DNNs, most of the adversarial methods such as FGSM, box-constrained L-BFGS, and ATN generate adversarial examples with small Lp-norm. However, these adversarial examples might contain many redundant perturbations. Removing these perturbations increases the quality of adversarial examples. Therefore, this paper proposes a method to improve the quality of adversarial examples by recognizing and then removing such perturbations. The proposed method includes two phases namely the autoencoder training phase and the improvement phase. In the autoencoder training phase, the proposed method trains an autoencoder that learns how to recognize redundant perturbations. In the second phase, the proposed method uses the trained autoencoder in combination with the greedy improvement step to produce more high-quality adversarial examples. The experiments on MNIST and CIFAR-10 have shown that the proposed method could improve the quality of adversarial examples significantly. In terms of L0-norm, the distance decreases by about 82%-95%. In terms of L2-norm, the distance drops by around 56%-81%. Additionally, the proposed method has a low computational cost. This shows the potential ability of the proposed method in practice.
Download

Paper Nr: 81
Title:

AnimaChaotic: AI-based Automatic Conversion of Children’s Stories to Animated 3D Videos

Authors:

Reem Abdel-Salam, Reem Gody, Mariam Maher, Hagar Hosny and Ahmed S. Kaseb

Abstract: Stories are an effective and entertaining way of teaching children about real-life experiences in an engaging way. Although many children’s stories are supplemented with graphical illustrations, having animated 3D video illustrations can effectively boost the learning process, especially for visual learners. However, producing animated 3D videos is a hard, expensive, and time-consuming process, so there is a need to automate this process. In this paper, we introduce AnimaChaotic, a story visualization system designed to automatically convert children’s short stories to animated 3D videos by leveraging Artificial Intelligence (AI) and computer graphics. Our Natural Language Processing (NLP) pipeline extracts visualizable information from the story such as actors and actions. Then, our object positioning algorithm determines the initial positions of the objects in the scene. Finally, the system animates the scene using different techniques including AI behaviors. A quantitative analysis of our system demonstrates a high precision and recall in extracting visualizable information. It also shows that our system outperforms existing solutions in terms of static scene generation. A qualitative analysis of the system shows that its output is visually acceptable and outperforms similar solutions.
Download

Paper Nr: 88
Title:

Post-hoc Global Explanation using Hypersphere Sets

Authors:

Kohei Asano and Jinhee Chun

Abstract: We propose a novel global explanation method for a pre-trained machine learning model. Generally, machine learning models behave as a black box. Therefore, developing a tool that reveals a model’s behavior is important. Some studies have addressed this issue by approximating a black-box model with another interpretable model. Although such a model summarizes a complex model, it sometimes provides incorrect explanations because of a gap between the complex model. We define hypersphere sets of two types that respectively approximate a model based on recall and precision metrics. A high-recall set of hyperspheres provides a summary of a black-box model. A high-precision one describes the model’s behavior precisely. We demonstrate from experimentation that the proposed method provides a global explanation for an arbitrary black-box model. Especially, it improves recall and precision metrics better than earlier methods.
Download

Paper Nr: 97
Title:

An Effective Method to Answer Multi-hop Questions by Single-hop QA System

Authors:

Kong Yuntao, Nguyen M. Phuong, Teeradaj Racharak, Tung Le and Nguyen L. Minh

Abstract: Multi-hop question answering (QA) requires a model to aggregate information from multiple paragraphs to predict the answer. Recent research on multi-hop QA has attempted this task by utilizing graph neural networks (GNNs) with sophisticated graph structures. While such models can achieve good performance, their computation is rather expensive. In this paper, we explore an alternative method that leverages a single-hop QA model to deal with multi-hop questions. Our system called ‘Answer Multi-hop questions by Single-hop QA’ (AMS) consists of three main parts that first filter a document and then conduct prediction using the attention-based single-hop QA model with multi-task learning. Specifically, AMS is constructed based on the co-attention and self-attention architecture. Lastly, consider that BERT-based model is pre-trained in a general domain and the data distribution can be different from multi-hop QA task. We propose two-step tuning mechanism to enhance the model’s performance, which is based on transfer learning from other QA datasets. To verify AMS effectiveness, we consider the previous state-of-the-art Hierarchical Graph Network (HGN) with the same document filter as our baseline. Experiments on HotpotQA show that AMS can outperform HGN by 1.78 points and 0.56 points for Joint EM and Joint F1, respectively. Meanwhile, it has smaller model’s size and uses less computational resource. We also experiment with other GNN-based models and achieve better results.
Download

Paper Nr: 100
Title:

Enhanced Local Gradient Smoothing: Approaches to Attacked-region Identification and Defense

Authors:

Cheng You-Wei and Wang Sheng-De

Abstract: Mainstream deep learning algorithms have been shown vulnerable to adversarial attacks - the deep models could be misled by adding small unnoticeable perturbations to the original input image. These attacks could pose security challenges in real-world applications. The paper focuses on how to defend against an adversarial patch attack that confines such noises within a small and localized patch area. We will discuss how an adversarial sample affects the classifier output from the perspective of a deep model by visualizing its saliency map. On the basis of our baseline method: Local Gradients Smoothing, we further design two methods called Saliency-map-based Local Gradients Smoothing and Weighted Local Gradients Smoothing, integrating saliency maps with local gradient maps to accurately locate a possible attacked region and perform smoothing accordingly. Experimental results show that our proposed method could reduce the probability of false smoothing and increase the overall accuracy significantly.
Download

Paper Nr: 101
Title:

GAN-based Intrinsic Exploration for Sample Efficient Reinforcement Learning

Authors:

Doğay Kamar, Nazím K. Üre and Gözde Ünal

Abstract: In this study, we address the problem of efficient exploration in reinforcement learning. Most common exploration approaches depend on random action selection, however these approaches do not work well in environments with sparse or no rewards. We propose Generative Adversarial Network-based Intrinsic Reward Module that learns the distribution of the observed states and sends an intrinsic reward that is computed as high for states that are out of distribution, in order to lead agent to unexplored states. We evaluate our approach in Super Mario Bros for a no reward setting and in Montezuma’s Revenge for a sparse reward setting and show that our approach is indeed capable of exploring efficiently. We discuss a few weaknesses and conclude by discussing future works.
Download

Paper Nr: 107
Title:

Aerial Fire Image Synthesis and Detection

Authors:

Sandro Campos and Daniel C. Silva

Abstract: Unmanned Aerial Vehicles appear as efficient platforms for fire detection and monitoring due to their low cost and flexibility features. Detecting flames and smoke from above is performed visually or by employing onboard temperature and gas concentration sensors. However, approaches based on computer vision and machine learning techniques have identified a pertinent problem of class imbalance in the fire image domain, which hinders detection performance. To represent fires visually and in an automated fashion, a residual neural network generator based on CycleGAN is implemented to perform unpaired image-to-image translation of non-fire images obtained from Bing Maps to the fire domain. Additionally, the adaptation of ERNet, a lightweight disaster classification network trained on the real fire domain, enables simulated aircraft to carry out fire detection along their trajectories. We do so under an environment comprised of a multi-agent distributed platform for aircraft and environmental disturbances, which helps tackle the previous inconvenience by accelerating artificial aerial fire imagery acquisition. The generator was tested using the metric of Fréchet Inception Distance, and qualitatively, resorting to the opinion of 122 subjects. The images were considered diverse and of good quality, particularly for the forest and urban scenarios, and their anomalies were highlighted to identify further improvements. The detector performance was evaluated in interaction with the simulation platform. It was proven to be compatible with real-time requirements, processing detection requests at around 100 ms, reaching an accuracy of 90.2% and a false positive rate of 4.5%.
Download

Paper Nr: 124
Title:

Improving Social Emotion Prediction with Reader Comments Integration

Authors:

Abdullah Alsaedi, Phillip Brooker, Floriana Grasso and Stuart Thomason

Abstract: Social emotion prediction is concerned with the prediction of the reader’s emotion when exposed to a text. In this paper, we propose a comment integration method for social emotion prediction. The basic intuition is that enriching social media posts with related comments can enhance the models’ ability to capture the conversation context, and hence improve the performance of social emotion prediction. We developed three models that use the comment integration method with different approaches: word-based, topic-based, and deep learning-based. Results show that our proposed models outperform popular models in terms of accuracy and F1-score.
Download

Paper Nr: 126
Title:

Towards Circular and Asymmetric Cooperation in a Multi-player Graph-based Iterated Prisoner’s Dilemma

Authors:

Tangui L. Gléau, Xavier Marjou, Tayeb Lemlouma and Benoit Radier

Abstract: In collaborations involving multiple actors, it is well known that tensions between individual interest and global welfare can emerge: actors are personally incentivized to have selfish behavior whereas mutual cooperation may provide a better outcome for all. Known as social dilemmas, these cooperation paradigms have aroused renewed interest in solving social issues, particularly in environmental and energy issues. Hybrids methods with Reinforcement Learning (RL) policies and Tit-for-Tat (TFT) strategies have proven successful to identify fruitful collaboration in complex social dilemmas. However, there are also many situations, where cooperation cannot always be given back directly, and has instead to be carried out through one or more intermediary actor(s). This specificity ruins win-win approaches like TFT. To address this specificity, we introduce a Graph-based Iterated Prisoner’s Dilemma: a N-player game in which the possible cooperation between players is modeled by a weighted directed graph. In addition to this new paradigm, we propose a graph-based TFT algorithm that we evaluate on multiple scenarios and compare to other algorithms. Our experiments show that leveraging a graph-based structure in the original TFT algorithm allows it to spread favor better collaboration synergies in most situations.
Download

Paper Nr: 128
Title:

Unsupervised Activity Recognition Using Trajectory Heatmaps from Inertial Measurement Unit Data

Authors:

Orhan Konak, Pit Wegner, Justin Albert and Bert Arnrich

Abstract: The growth of sensors with varying degrees of integration and functionality has inevitably led to their entry into various fields such as digital health. Here, sensors that can record acceleration and rotation rates, so- called Inertial Measurement Units (IMU), are primarily used to distinguish between different activities, also known as Human Activity Recognition (HAR). If the associations of the motion data to the activities are not known, clustering methods are used. There are many algorithmic approaches to identify similarity structures in the incoming sensor data. These differ mainly in their notion of similarity and grouping, as well as in their complexity. This work aimed to investigate the impact of transforming the incoming time-series data into corresponding motion trajectories and trajectory heatmap images before forwarding it to well-known clustering models. All three input variables were given to the same clustering algorithms, and the results were compared using different evaluation metrics. This work shows that transforming sensor data into trajectories and images leads to a significant increase in cluster assignment for all considered models and different metrics.
Download

Paper Nr: 129
Title:

Detecting Narcissist Dark Triad Psychological Traits from Twitter

Authors:

Lidice Haz, Miguel Á. Rodríguez-García and Alberto Fernández

Abstract: The fundamental basis of human behavior is personality. This human characteristic influences the tastes and preferences of individuals and the way they interact and communicate with each other. Nowadays, individuals express their feelings, opinions, and ideas to the world by using digital communication platforms such as social media. In recent years, several studies have proposed different models that apply Artificial Intelligence techniques to identify personality traits based mainly on the Big 5 Personality Model and the three subpersonalities of the Dark Triad through the linguistic analysis of their comments online. In this work, we present a study about identifying narcissist dark triad psychological traits. We propose two Machine Learning models to analyze users’ behavior in social media. Concretely, we develop a Support Vector Machine and Naïve Bayes method to classify the comments as having non-narcissist or narcissist traits. To train and test the developed method, we have employed NLP techniques to process comments from Twitter and created a manual dataset. Three different techniques have been designed and applied to label each tweet and comment. Then, we conducted several evaluations in which both models reached promising results.
Download

Paper Nr: 130
Title:

Post-hoc Diversity-aware Curation of Rankings

Authors:

Vassilis Markos and Loizos Michael

Abstract: We consider the problem of constructing rankings that exhibit a prescribed level of diversity. We take a black- box view of the ranking procedure itself, and choose, instead, to post-hoc curate any given ranking, deviating from the original ranking only to the extent allowed by any given constraints. The curation algorithm that we present is oblivious to how diversity is measured, and returns shuffled versions of the original rankings that are optimal in terms of their exhibited level of diversity. Our empirical evaluation on synthetic data demonstrates the effectiveness and efficiency of our methodology, across a number of diversity metrics from the literature.
Download

Paper Nr: 131
Title:

Developing a Successful Bomberman Agent

Authors:

Dominik Kowalczyk, Jakub Kowalski, Hubert Obrzut, Michał Maras, Szymon Kosakowski and Radosław Miernik

Abstract: In this paper, we study AI approaches to successfully play a 2–4 players, full information, Bomberman variant published on the CodinGame platform. We compare the behavior of three search algorithms: Monte Carlo Tree Search, Rolling Horizon Evolution, and Beam Search. We present various enhancements leading to improve the agents’ strength that concern search, opponent prediction, game state evaluation, and game engine encoding. Our top agent variant is based on a Beam Search with low-level bit-based state representation and evaluation function heavy relying on pruning unpromising states based on simulation-based estimation of survival. It reached the top one position among the 2,300 AI agents submitted on the CodinGame arena.
Download

Paper Nr: 133
Title:

A Hierarchical Probabilistic Divergent Search Applied to a Binary Classification

Authors:

Senthil Murugan, Enrique Naredo, Douglas M. Dias, Conor Ryan, Flaviano Godinez and James V. Patten

Abstract: The trend in recent years of the scientific community on solving a wide range of problems through Artificial Intelligence has highlighted the benefits of open-ended search algorithms. In this paper we apply a probabilistic version for a divergent search algorithm in combination of a strategy to reduce the number of evaluations and computational effort by gathering the population from a Genetic Programming algorithm into groups and pruning the worst groups each certain number of generations. The combination proposed has shown encouraging results against a standard GP implementation on three binary classification problems, where the time taken to run an experiment is significantly reduced to only 5% of the total time from the standard approach while still maintaining, and indeed exceeding in the experimental results.
Download

Paper Nr: 140
Title:

Vectorization of Bias in Machine Learning Algorithms

Authors:

Sophie Bekerman, Eric Chen, Lily Lin and George D. Monta Nez

Abstract: We develop a method to measure and compare the inductive bias of classifications algorithms by vectorizing aspects of their behavior. We compute a vectorized representation of the algorithm’s bias, known as the inductive orientation vector, for a set of algorithms. This vector captures the algorithm’s probability distribution over all possible hypotheses for a classification task. We cluster and plot the algorithms’ inductive orientation vectors to visually characterize their relationships. As algorithm behavior is influenced by the training dataset, we construct a Benchmark Data Suite (BDS) matrix that considers algorithms’ pairwise distances across many datasets, allowing for more robust comparisons. We identify many relationships supported by existing literature, such as those between k-Nearest Neighbor and Random Forests and among tree-based algorithms, and evaluate the strength of those known connections, showing the potential of this geometric approach to investigate black-box machine learning algorithms.
Download

Paper Nr: 147
Title:

Automatic Arabic Poem Generation with GPT-2

Authors:

Mohamed El Ghaly Beheitt and Moez Ben Haj Hmida

Abstract: Automatically generating poetry by computers is a challenging topic that requires the use of advanced deep learning techniques. While much attention has been given to English and Chinese poem generation, there are few significant efforts considering other languages. Generating poems in Arabic is a difficult task due to the complexity of the Arabic language grammatical structure. In this paper, we investigate the feasibility of training generative pre-trained language model GPT-2 to generate Arabic poems. The results of the experiments, which included the BLEU score as well as human assessments, confirmed the effectiveness of our proposed model. Both automatic and human evaluations show that our proposed model outperforms existing models in generating Arabic poetry.
Download

Paper Nr: 148
Title:

X-Ray Classification to Detect COVID-19 using Ensemble Model

Authors:

Ishmam A. Solaiman, Tasnim I. Sanjana, Samila Sobhan, Tanzila S. Maria and Md. K. Rahman

Abstract: Diagnosis with medical images has soared to new heights and play massive roles in assisting radiologists to detect and analyse medical conditions. Computer-Aided Diagnosis systems are successfully used to detect tuberculosis, pneumonia, etc. from CXR images. CNNs have been adopted by many studies and achieved laudable results in the field of medical image diagnosis, having attained state-of-art performance by training on labeled data. This paper aims to propose an Ensemble model using a combination of deep CNN architectures, which are Xception, InceptionResnetV2, VGG19, DenseNet-201, and NasNetLarge, using image processing and artificial intelligence algorithms to quickly and accurately identify COVID-19 and other coronary diseases from X-Rays to stop the rapid transmission of the virus. We have used classifiers for the Xception model, VGG19, and InceptionResnet model and compiled a CXR dataset from various open datasets. Since the dataset lacked 1000 viral pneumonia images , we used image augmentation and focal loss to compensate for the unbalanced data and to introduce more variation. After implementing the focal loss function, we got better results. Moreover, we implemented transfer learning using ImageNet weights. Finally, we obtained a training accuracy of 92% to 94% across all models. Our accuracy of the Ensemble Model was 96.25%.
Download

Paper Nr: 161
Title:

On-the-Fly Knowledge Acquisition for Automated Planning Applications: Challenges and Lessons Learnt

Authors:

Saumya Bhatnagar, Sumit Mund, Enrico Scala, Keith McCabe, Thomas L. McCluskey and Mauro Vallati

Abstract: Automated planning is a prominent AI challenge, and it is now exploited in a range of real-world applications. There are three crucial aspects of automated planning: the planning engine, the domain model, and the problem instance. While the planning engine and the domain model can be engineered and optimised offline, in many applications there is the need to generate problem instances on the fly. In this paper we focus on the challenges of on-the-fly knowledge acquisition for complex and variegated problem instances. We consider as a case study the application of planning to urban traffic control and we describe the designed and developed knowledge acquisition process. This allows us to discuss a range of lessons learned from the experience, and to point to important lines of research to support the knowledge acquisition process for automated planning applications.
Download

Paper Nr: 165
Title:

Univariate Time Series Prediction using Data Stream Mining Algorithms and Temporal Dependence

Authors:

Marcos Alberto Mochinski, Jean Paul Barddal and Fabrício Enembreck

Abstract: In this paper, we present an exploratory study conducted to evaluate the impact of temporal dependence modeling on time series forecasting with Data Stream Mining (DSM) techniques. DSM algorithms have been used successfully in many domains that exhibit continuous generation of non-stationary data. However, the use of DSM in time series is rare since they usually are univariate and exhibit strong temporal dependence. This is the main motivation for this work, such that this study mitigates such gap by presenting a univariate time series prediction method based on AdaGrad (a DSM algorithm), Auto.Arima (a statistical method) and features extracted from adjusted autocorrelation function (ACF) coefficients. The proposed method uses adjusted ACF features to convert the original series observations into multivariate data, executes the fitting process using the DSM and the statistical algorithm, and combines the AdaGrad's and Auto.Arima's forecasts to establish the final predictions. Experiments conducted with five datasets containing 141,558 time series resulted in up to 12.429% improvements in sMAPE (Symmetric Mean Average Percentage Error) error rates when compared to Auto.Arima. The results depict that combining DSM with ACF features and statistical time series methods is a suitable approach for univariate forecasting.
Download

Paper Nr: 169
Title:

Optimizing Multi-objective Knapsack Problem using a Hybrid Ant Colony Approach within Multi Directional Framework

Authors:

Imen Ben Mansour

Abstract: Balancing the convergence and diversity simultaneously is very challenging for multi-objective evolutionary algorithms on solving multi-objective optimization problems (MOPs). The proposed approach MD-HACO coupled an Ant Colony Optimization (ACO) algorithm with a multi-objective local search procedure, and evolves it into a multi-directional framework. The idea is to optimize the overall quality of Pareto set approximation by using different configurations of the hybrid approach by means of different directional vectors. During the optimization process, the artificial ants work in different search directions in the objective space trying to approximate small parts of the Pareto front. Afterward, a local search procedure is applied to each sub-region to enhance the search process toward the extreme Pareto-optimal solutions with respect to the weight vector under consideration. A multi-directional set holding the non-dominated solutions according to all directional archives is maintained. The proposed approach is tested on widely used multi-objective multi-dimensional knapsack problem (MOMKP) instances and compared with well-known state-of-the-art algorithms. Experiments highlight that the use of a multi-directional paradigm as well as a hybrid schema can lead to interesting results on the MOMKP and ensure a good balance between convergence and diversity.
Download

Paper Nr: 177
Title:

Towards a Certification of Deep Image Classifiers against Convolutional Attacks

Authors:

Mallek Mziou-Sallami and Faouzi Adjed

Abstract: Deep learning models do not achieve sufficient confidence, explainability and transparency levels to be integrated into safety-critical systems. In the context of DNN-based image classifier, robustness have been first studied under simple image attacks (2D rotation, brightness), and then, subsequently, under other geometrical perturbations. In this paper, we intend to introduce a new method to certify deep image classifiers against convolutional attacks. Using the abstract interpretation theory, we formulate the lower and upper bounds with abstract intervals to support other classes of advanced attacks including image filtering. We experiment the proposed method on MNIST and CIFAR10 databases and several DNN architectures. The obtained results show that convolutional neural networks are more robust against filtering attacks. Multilayered perceptron robustness decreases when increasing number of neurons and hidden layers. These results prove that the complexity of DNN models improves prediction’s accuracy but often impacts robustness.
Download

Paper Nr: 184
Title:

FOREAL: RoBERTa Model for Fake News Detection based on Emotions

Authors:

Vladislav Kolev, Gerhard Weiss and Gerasimos Spanakis

Abstract: Detecting false information in the form of fake news has become a bigger challenge than anticipated. There are multiple promising ways of approaching such a problem, ranging from source-based detection, linguistic feature extraction, and sentiment analysis of articles. While analyzing the sentiment of text has produced some promising results, this paper explores a rather more fine-grained strategy of classifying news as fake or real, based solely on the emotion profile of an article’s title. A RoBERTa model was first trained to perform Emotion Classification, achieving test accuracy of about 90%. Six basic emotions were used for the task, based on the prominent psychologist Paul Ekman - fear, joy, anger, sadness, disgust and surprise. A seventh emotional category was also added to represent neutral text. Model performance was also validated by comparing classification results to other state-of-the-art models, developed by other groups. The model was then used to make inference on the emotion profile of news titles, returning a probability vector, which describes the emotion that the title conveys. Having the emotion probability vectors for each article’s title, another Binary Random Forest classifier model was trained to evaluate news as either fake or real, based solely on their emotion profile. The model achieved up to 88% accuracy on the Kaggle Fake and Real News Dataset, showing there is a connection present between the emotion profile of news titles and if the article is fake or real.
Download

Paper Nr: 185
Title:

Parsimonious Representation of Knowledge Uncertainty using Metadata about Validity and Completeness

Authors:

Célia C. Pereira, Didier Dubois, Henri Prade and Andrea B. Tettamanzi

Abstract: We investigate how metadata about the uncertainty of knowledge contained in a knowledge base can be expressed parsimoniously and used for reasoning. We propose an approach based on possibility theory, whereby a classical knowledge base plus metadata about the degree of validity and completeness of some of its portions are used to represent a possibilistic belief base. We show how reasoning on such belief base can be done using a classical reasoner.
Download

Paper Nr: 196
Title:

Enabling Markovian Representations under Imperfect Information

Authors:

Francesco Belardinelli, Borja G. León and Vadim Malvone

Abstract: Markovian systems are widely used in reinforcement learning (RL), when the successful completion of a task depends exclusively on the last interaction between an autonomous agent and its environment. Unfortunately, real-world instructions are typically complex and often better described as non-Markovian. In this paper we present an extension method that allows solving partially-observable non-Markovian reward decision processes (PONMRDPs) by solving equivalent Markovian models. This potentially facilitates Markovian-based state-of-the-art techniques, including RL, to find optimal behaviours for problems best described as PONMRDP. We provide formal optimality guarantees of our extension methods together with a counterexample illustrating that naive extensions from existing techniques in fully-observable environments cannot provide such guarantees.
Download

Paper Nr: 198
Title:

T-Balance: A Unified Mechanism for Taxi Scheduling in a City-scale Ride-sharing Service

Authors:

Jiyao Li and Vicki H. Allan

Abstract: In this paper, we propose a unified mechanism known as T-Balance for scheduling taxis across a city. Balancing the supplies and demands in a city scale is a challenging problem in the field of the ride-sharing service. To tackle the problem, we design a unified mechanism considering two important processes in ride-sharing service: ride-matching and vacant taxi repositioning. For rider-matching, the Scoring Ride-matching with Lottery Selection (SRLS) is proposed. With the help of Lottery Selection (LS) and smoothed popularity score, the Scoring Ride-matching with Lottery Selection (SRLS) can balance supplies and demands well, both in the local neighborhood areas and hot places across the city. In terms of vacant taxi repositioning, we propose Qlearning Idle Movement (QIM) to direct vacant taxis to the most needed places in the city, adapting to dynamic change environments. The experimental results verify that the unified mechanism is effective and flexible.
Download

Paper Nr: 199
Title:

Jason Agents for Knowledge-aware Information Retrieval Filters

Authors:

Dima El Zein and Célia C. Pereira

Abstract: This paper proposes a novel use of Belief-Desire-Intention agents in Information Retrieval. We present a cognitive agent that builds its beliefs about the user’s knowledge during his/her interaction with the search system. The agent reasons about those beliefs and derives new ones using contextual rules. Those beliefs serve to personalise the search results accordingly. The framework is developed using an extended version of the Jason agent programming language; the choice of Jason’s extension to model the agents is justified by some of its advantageous features, in particular, the possibility to represent gradual beliefs. A running example will illustrate the presented work and highlight its added value.
Download

Paper Nr: 204
Title:

Robust Traffic Signal Timing Control using Multiagent Twin Delayed Deep Deterministic Policy Gradients

Authors:

Priya Shanmugasundaram and Shalabh Bhatnagar

Abstract: Traffic congestion is an omnipresent and serious problem that impacts people around the world on a daily basis. It requires solutions that can adapt to the changing traffic environments and reduce traffic congestion not only across local intersections but also across the global road network. Traditional traffic control strategies suffer from being too simplistic and moreover, they cannot scale to real-world dynamics. Multiagent reinforcement learning is being widely researched to develop intelligent transportation systems where the different intersections on a road network co-operate to ease vehicle delay and traffic congestion. Most of the literature on using Multiagent reinforcement learning methods for traffic signal control is focussed on applying multi-agent Q learning and discrete-action based control methods. In this paper, we propose traffic signal control using Multiagent Twin Delayed Deep Deterministic Policy Gradients (MATD3). The proposed control strategy is evaluated by exposing it to different time-varying traffic flows on simulation of road networks created on the traffic simulation platform SUMO. We observe that our method is robust to the different kinds of traffic flows and consistently outperforms the state-of-the-art counterparts by significantly reducing the average vehicle delay and queue length.
Download

Paper Nr: 206
Title:

Identifying Soft Cores in Propositional Formulæ

Authors:

Gilles Audemard, Jean-Marie Lagniez, Marie Miceli and Olivier Roussel

Abstract: In view of the emergence of explainable AI, many new concepts intend to explain why systems exhibit certain behaviors while other behaviors are excluded. When dealing with constraints, explanations can take the form of subsets having few solutions, while being sufficiently small for ensuring that they are intelligible enough. To make it formal, we present a new notion, called soft core, characterizing both small and highly constrained parts of GCNF instances, whether satisfiable or not. Soft cores can be used in unsatisfiable instances as an alternative to MUSes (Minimal Unsatisfiable Subformulæ) or in satisfiable ones as an alternative to MESes (Minimal Equivalent Subformulæ). We also provide an encoding to translate soft cores instances into MAX#SAT instances. Finally, we propose a new method to solve MAX#SAT instances and we use it to extract soft cores.
Download

Paper Nr: 207
Title:

Adversarial Examples by Perturbing High-level Features in Intermediate Decoder Layers

Authors:

Vojtěch Čermák and Lukáš Adam

Abstract: We propose a novel method for creating adversarial examples. Instead of perturbing pixels, we use an encoder-decoder representation of the input image and perturb intermediate layers in the decoder. This changes the high-level features provided by the generative model. Therefore, our perturbation possesses semantic meaning, such as a longer beak or green tints. We formulate this task as an optimization problem by minimizing the Wasserstein distance between the adversarial and initial images under a misclassification constraint. We employ the projected gradient method with a simple inexact projection. Due to the projection, all iterations are feasible, and our method always generates adversarial images. We perform numerical experiments by fooling MNIST and ImageNet classifiers in both targeted and untargeted settings. We demonstrate that our adversarial images are much less vulnerable to steganographic defence techniques than pixel-based attacks. Moreover, we show that our method modifies key features such as edges and that defence techniques based on adversarial training are vulnerable to our attacks.
Download

Paper Nr: 211
Title:

Cognitive Planning in Motivational Interviewing

Authors:

Emiliano Lorini, Nicolas Sabouret, Brian Ravenet, Jorge Fernandez and Céline Clavel

Abstract: This paper presents a cognitive planning model that implements the principles of motivational interviewing, a counseling method used to guide people in adopting behavior changes. This planning system is part of a wider dialogical architecture of artificial counseling agent. We present the formal model and planning problem. We show how it can be used to plan for dialogue in the architecture. We illustrate its functionalities on a simple example.
Download

Paper Nr: 226
Title:

Developing and Experimenting on Approaches to Explainability in AI Systems

Authors:

Yuhao Zhang, Kevin McAreavey and Weiru Liu

Abstract: There has been a sharp rise in research activities on explainable artificial intelligence (XAI), especially in the context of machine learning (ML). However, there has been less progress in developing and implementing XAI techniques in AI-enabled environments involving non-expert stakeholders. This paper reports our investigations into providing explanations on the outcomes of ML algorithms to non-experts. We investigate the use of three explanation approaches (global, local, and counterfactual), considering decision trees as a use case ML model. We demonstrate the approaches with a sample dataset, and provide empirical results from a study involving over 200 participants. Our results show that most participants have a good understanding of the generated explanations.
Download

Paper Nr: 227
Title:

The Gopher Grounds: Testing the Link between Structure and Function in Simple Machines

Authors:

Anshul Kamath, Jiayi Zhao, Nick Grisanti and George D. Montañez

Abstract: Does structure dictate function and can function be reliably inferred from structure? Previous work has shown that an artificial agent’s ability to detect function (e.g., lethality) from structure (e.g., the coherence of traps) can confer measurable survival advantages. We explore the link between structure and function in simple combinatorial machines, using genetic algorithms to generate traps with structure (coherence) and no function (no lethality), generate traps with function and no structure, and generate traps with both structure and function. We explore the characteristics of the algorithmically generated traps, examine the genetic algorithms’ ability to produce structure, function, and their combination, and investigate what resources are needed for the genetic algorithms to reliably succeed at these tasks. We find that producing lethality (function) is easier than producing coherence (structure) and that optimizing for one does not reliably produce the other.
Download

Paper Nr: 228
Title:

Uncertainty Guided Pseudo-Labeling: Estimating Uncertainty on Ambiguous Data for Escalating Image Recognition Performance

Authors:

Kyung H. Park and HyunHee Chung

Abstract: Upon the dominant accomplishments of deep neural networks, recent studies have scrutinized a robust model under the inherently ambiguous samples. Prior works have achieved superior performance under these ambiguous samples through label distribution approaches, assuming the existence of multiple human annotators. However, the aforementioned problem setting is not generally feasible due to resource constraints. For a generally applicable solution to the ambiguity problem, we propose Uncertainty-Guided Pseudo-Labeling (UGPL), a proof-of-concept level framework that leverages ambiguous samples on elevating the image recognition performance. Key contributions of our study are as follows. First, we constructed synthetic ambiguous datasets as there were no public benchmark dataset that deals with ambiguity problem. Given ambiguous samples, we empirically showed that not every ambiguous sample has meaningful knowledge consistent to the obvious samples at the target classes. We then examined uncertainty can be a possible proxy for measuring the effectiveness of ambiguous sample’s knowledge toward the escalation of image recognition performance. Moreover, we validated pseudo-labeled ambiguous samples with low uncertainty better contributes to the test accuracy elevation. Lastly, we validated the UGPL showed larger accuracy elevation under the small size of obvious samples; thus, general practitioners can be widely benefited. To this end, we suggest further avenues of improvement practical techniques that resolve the ambiguity problem.
Download

Paper Nr: 232
Title:

Graph Convolutional Networks for Turn-Based Strategy Games

Authors:

Wanxiang Li, Houkuan He, Chu-Hsuan Hsueh and Kokolo Ikeda

Abstract: In this paper, we research Turn-Based Strategy (TBS) games that allow players to move multiple pieces in one turn and have multiple initial states. Compared to a game like Chess, which allows only one piece to move per turn and has a single initial state, it is difficult to create a strong computer player for such a group of TBS games. Deep learning methods such as AlphaZero and DQN are often used to create strong computer players. Convolutional neural networks (CNNs) are used to output policies and/or values, and input states are represented as “image”-like data. For TBS games, we consider that the relationships among units are more important than their absolute positions, and we attempt to represent the input states as “graphs”. In addition, we adopt graph convolutional neural networks (GCNs) as the suitable networks when inputs are graphs. In this research, we use a TBS game platform TUBSTAP as our test game and propose to (1) represent TUBSTAP game states as graphs, (2) employ GCNs as value network to predict the game result (win/loss/tie) by supervised learning, (3) compare the prediction accuracy of GCNs and CNNs, and (4) compare the playing strength of GCNs and CNNs when the learned value network is incorporated into a tree search. Experimental results show that the combination of graph input and GCN improves the accuracy of predicting game results and the strength of playing TUBSTAP.
Download

Paper Nr: 233
Title:

Time to Focus: A Comprehensive Benchmark using Time Series Attribution Methods

Authors:

Dominique Mercier, Jwalin Bhatt, Andreas Dengel and Sheraz Ahmed

Abstract: In the last decade neural network have made huge impact both in industry and research due to their ability to extract meaningful features from imprecise or complex data, and by achieving super human performance in several domains. However, due to the lack of transparency the use of these networks is hampered in the areas with safety critical areas. In safety-critical areas, this is necessary by law. Recently several methods have been proposed to uncover this black box by providing interpreation of predictions made by these models. The paper focuses on time series analysis and benchmark several state-of-the-art attribution methods which compute explanations for convolutional classifiers. The presented experiments involve gradient-based and perturbation-based attribution methods. A detailed analysis shows that perturbation-based approaches are superior concerning the Sensitivity and occlusion game. These methods tend to produce explanations with higher continuity. Contrarily, the gradient-based techniques are superb in runtime and Infidelity. In addition, a validation the dependence of the methods on the trained model, feasible application domains, and individual characteristics is attached. The findings accentuate that choosing the best-suited attribution method is strongly correlated with the desired use case. Neither category of attribution methods nor a single approach has shown outstanding performance across all aspects.
Download

Paper Nr: 235
Title:

Grouping of Maintenance Actions with Deep Reinforcement Learning and Graph Convolutional Networks

Authors:

David Kerkkamp, Zaharah A. Bukhsh, Yingqian Zhang and Nils Jansen

Abstract: Reinforcement learning (RL) has shown promising performance in several applications such as robotics and games. However, the use of RL in emerging real-world domains such as smart industry and asset management remains scarce. This paper addresses the problem of optimal maintenance planning using historical data. We propose a novel Deep RL (DRL) framework based on Graph Convolutional Networks (GCN) to leverage the inherent graph structure of typical assets. As demonstrator, we employ an underground sewer pipe network. In particular, instead of dispersed maintenance actions of individual pipes across the network, the GCN ensures the grouping of maintenance actions of geographically close pipes. We perform experiments using the distinct physical characteristics, deterioration profiles, and historical data of sewer inspections within an urban environment. The results show that combining Deep Q-Networks (DQN) with GCN leads to structurally more reliable networks and a higher degree of maintenance grouping, compared to DQN with fully-connected layers and standard preventive and corrective maintenance strategy that are often adopted in practice. Our approach shows potential for developing efficient and practical maintenance plans in terms of cost and reliability.
Download

Paper Nr: 242
Title:

Parameter Setting in SAT Solver using Machine Learning Techniques

Authors:

Filip Beskyd and Pavel Surynek

Abstract: Boolean satisfiability (SAT) solvers are essential tools for many domains in computer science and engineering. Modern complete search-based SAT solvers represent a universal problem solving tool which often provide higher efficiency than ad-hoc direct solving approaches. Over the course of at least two decades of SAT related research, many variable and value selection heuristics were devised. Heuristics can usually be tuned by single or multiple numerical parameters prior to executing the search process over the concrete SAT instance. In this paper we present a machine learning approach that predicts the parameters of heuristic from the underlying structure of the input SAT instance.
Download

Paper Nr: 263
Title:

Intelligent Diagnosis of Breast Cancer with Thermograms using Convolutional Neural Networks

Authors:

Nurduman Aidossov, Aigerim Mashekova, Yong Zhao, Vasilios Zarikas, Eddie K. Ng and Olzhas Mukhmetov

Abstract: Breast cancer is a serious public health issue among women all over the world. The main methods of breast cancer diagnosis include ultrasound, mammography and Magnetic Resonance Imaging (MRI). However, the existing methods of diagnosis are not appropriate for regular mass screening in short intervals. On the other hand, there is one non-invasive and low-cost method for mass and regular screening which is the so-called thermography. Recent studies show rapid quality improvement of thermal cameras as well as distinct development of machine learning techniques that can be combined together to enhance the technology of breast cancer detection. Machine learning technologies can potentially be used to support the interpretation of thermal images and help physicians to automatically determine the locations and sizes of tumors, blood perfusion, and other patient-specific properties of breast tissues. In this study, we aim to develop CNN techniques for intelligent precision breast tumor diagnosis. The main innovation of our work is the use of breast thermograms from a multicenter database without preprocessing for binary classification. The results presented in this paper highlight the usefulness and efficiency of deep learning for standardized analysis of thermograms. It is found that the model developed can have an accuracy of 80.77%, sensitivity of 44.44 % and the specificity of 100%.
Download

Paper Nr: 271
Title:

Understanding Summaries: Modelling Evaluation in Extractive Summarisation Techniques

Authors:

Victor Margallo

Abstract: In the task of providing extracted summaries, the assessment of quality evaluation has been traditionally tackled with n-gram, word sequences, and word pairs overlapping metrics with human annotated summaries for theoretical benchmarking. This approach does not provide an end solution for extractive summarising algorithms as output summaries are not evaluated for new texts. Our solution proposes the expansion of a graph extraction method together with an understanding layer before delivering the final summary. With this technique we strive to achieve a categorisation of acceptable output summaries. Our understanding layer judges correct summaries with 91% accuracy and is in line with experts’ labelling providing a strong inter-rater reliability (0.73 Kappa statistic).
Download

Paper Nr: 274
Title:

eSardine: A General Purpose Platform with Autonomous AI and Explainable Outputs

Authors:

Inês R. Lima, Nuno Leite, Adriano Pinto, Pedro Pires, Carlos Martins and Nuno V. Lopes

Abstract: The combination of high computational power and data awareness triggered an increasing demand for business applications from industrial players. However, harnessing the knowledge from data requires expertise, usually being a time-consuming task. Additionally, the users’ trust in the results obtained is commonly compromised due to the black box behavior of most Machine Learning models. This paper proposes a general-purpose platform, eSardine, that leverages automatic machine learning and explainability to produce fast, reliable, and interpretable results. The eSardine platform integrates forefront tools to enhance, and automate the data science process, with minimal human interaction. For any tabular supervised classification and regression problems, predicted outputs are given, as well as an explainability report of each prediction. The inclusion of AutoML tools, i.e. , automatic model tuning and selection, presented a strong baseline whose capabilities are amplified by built-in, yet customizable, autonomous processing mechanisms. The explainable reports aim to increase users’ confidence in the models’ quality and robustness. Furthermore, in the industrial context, understanding key factors unveiled in these reports is determinant to increase the business model’s profitability. The platform was evaluated in two public datasets, where it outperformed state-of-the-art results.
Download

Paper Nr: 278
Title:

Predicting the Intended Action using Internal Simulation of Perception

Authors:

Zahra Gharaee

Abstract: This article proposes an architecture, which allows the prediction of intention by internally simulating perceptual states represented by action pattern vectors. To this end, associative self-organising neural networks (A-SOM) is utilised to build a hierarchical cognitive architecture for recognition and simulation of the skeleton based human actions. The abilities of the proposed architecture in recognising and predicting actions is evaluated in experiments using three different datasets of 3D actions. Based on the experiments of this article, applying internally simulated perceptual states represented by action pattern vectors improves the performance of the recognition task in all experiments. Furthermore, internal simulation of perception addresses the problem of having limited access to the sensory input, and also the future prediction of the consecutive perceptual sequences. The performance of the system is compared and discussed with similar architecture using self-organizing neural networks (SOM).
Download

Paper Nr: 280
Title:

Towards More Reliable Text Classification on Edge Devices via a Human-in-the-Loop

Authors:

Jakob S. Andersen and Olaf Zukunft

Abstract: Reliably classifying huge amounts of textual data is a primary objective of many machine learning applications. However, state-of-the-art text classifiers require extensive computational resources, which limit their applicability in real-world scenarios. In order to improve the application of lightweight classifiers on edge devices, e.g. personal work stations, we adapt the Human-in-the-Loop paradigm to improve the accuracy of classifiers without re-training by manually validating and correcting parts of the classification outcome. This paper performs a series of experiments to empirically assess the performance of the uncertainty-based Human-in-the-Loop classification of nine lightweight machine learning classifiers on four real-world classification tasks using pre-trained SBERT encodings as text features. Since time efficiency is crucial for interactive machine learning pipelines, we further compare the training and inference time to enable rapid interactions. Our results indicate that lightweight classifiers with a human in the loop can reach strong accuracies, e.g. improving a classifier’s F1-Score from 90.19 to 97% when 22.62% of a dataset is classified manually. In addition, we show that SBERT based classifiers are time efficient and can be re-trained in < 4 seconds using a Logistic Regression model.
Download

Paper Nr: 283
Title:

Improved Predictive Fundamental Period Formula for Reinforced Concrete Structures through the Use of Machine Learning Algorithms

Authors:

Nathan Carstens, George Markou and Nikolaos Bakas

Abstract: With the development of technology and building materials, the world is moving towards creating a better and safer environment. One of the main challenges for reinforced concrete structures is the capability to withstand the seismic loads produced by earthquake excitations, through using the fundamental period of the structure. However, it is well documented that the current design formulae fail to predict the natural frequency of the considered structures due to their inability to incorporate the soil-structure interaction and other features of the structures. This research work extends a dataset containing 475 modal analysis results developed through a previous research work. The extended dataset was then used to develop three predictive fundamental period formulae using a machine learning algorithm that utilizes a higher-order, nonlinear regression modelling framework. The predictive formulae were validated with 60 out-of-sample modal analysis results. The numerical findings concluded that the fundamental period formulae proposed in this study possess superior prediction ability, compared to all other international proposed formulae, for the under-studied types of buildings.
Download

Paper Nr: 285
Title:

An Evolutionary-based Neural Network for Distinguishing between Genuine and Posed Anger from Observers’ Pupillary Responses

Authors:

Fan Wu, Md Rakibul Hasan and Md Zakir Hossain

Abstract: Future human-computing research could be enhanced by recognizing attitude/emotion (for example, anger) from observers’ reactions (for example, pupillary responses). This paper analyzes observers’ pupillary responses by developing neural network (NN) models to distinguish between genuine and posed anger. Any model’s relatively high classification accuracy means the pupillary responses and observed anger (genuine or posed) are deeply connected. In this connection, we implemented strategies for tuning parameters of the model, methods to optimize and compress the model structure, analyze the similarity of hidden units, and decide which of them should be removed. We achieved the goal of removing the network’s redundant neurons without significant performance decline and improved the training speed. Finally, our evolutionary-based NN model showed the highest accuracy of 86% with a 3-layers structure and outperformed the backpropagation- based NN. The high accuracy highlights the potential of our model to use in the future for distinguishing observers’ reactions to emotion/attitude recognition.
Download

Paper Nr: 286
Title:

Metaheuristics-based Exploration Strategies for Multi-Objective Reinforcement Learning

Authors:

Florian Felten, Grégoire Danoy, El-Ghazali Talbi and Pascal Bouvry

Abstract: The fields of Reinforcement Learning (RL) and Optimization aim at finding an optimal solution to a problem, characterized by an objective function. The exploration-exploitation dilemma (EED) is a well known subject in those fields. Indeed, a consequent amount of literature has already been proposed on the subject and shown it is a non-negligible topic to consider to achieve good performances. Yet, many problems in real life involve the optimization of multiple objectives. Multi-Policy Multi-Objective Reinforcement Learning (MPMORL) offers a way to learn various optimised behaviours for the agent in such problems. This work introduces a modular framework for the learning phase of such algorithms, allowing to ease the study of the EED in Inner-Loop MPMORL algorithms. We present three new exploration strategies inspired from the metaheuristics domain. To assess the performance of our methods on various environments, we use a classical benchmark - the Deep Sea Treasure (DST) - as well as propose a harder version of it. Our experiments show all of the proposed strategies outperform the current state-of-the-art ε-greedy based methods on the studied benchmarks.
Download

Paper Nr: 297
Title:

An Assessment of the Impact of OCR Noise on Language Models

Authors:

Konstantin Todorov and Giovanni Colavizza

Abstract: Neural language models are the backbone of modern-day natural language processing applications. Their use on textual heritage collections which have undergone Optical Character Recognition (OCR) is therefore also increasing. Nevertheless, our understanding of the impact OCR noise could have on language models is still limited. We perform an assessment of the impact OCR noise has on a variety of language models, using data in Dutch, English, French and German. We find that OCR noise poses a significant obstacle to language modelling, with language models increasingly diverging from their noiseless targets as OCR quality lowers. In the presence of small corpora, simpler models including PPMI and Word2Vec consistently outperform transformer-based models in this respect.
Download

Short Papers
Paper Nr: 4
Title:

Automatic Recognition of Human Activities Combining Model-based AI and Machine Learning

Authors:

Constantin Patsch, Marsil Zakour and Rahul Chaudhari

Abstract: Developing intelligent assistants for activities of daily living (ADL) is an important topic in eldercare due to the aging society in industrialized countries. Recognizing activities and understanding the human’s intended goal are the major challenges associated with such a system. We propose a hybrid model for composite activity recognition in a household environment by combining Machine Learning and knowledge-based models. The Machine Learning part, based on structural Recurrent Neural Networks (S-RNN), performs low-level activity recognition based on video data. The knowledge-based part, based on our extended Activation Spreading Network architecture, models and recognizes the contextual meaning of an activity within a plan structure. This model is able to recognize activities, underlying goals and sub-goals, and is able to predict subsequent activities. Evaluating our action S-RNN on data from the 3D activity simulator HOIsim yields a macro average F1 score of 0.97 and an accuracy of 0.99. The hybrid model is evaluated with activation value graphs.
Download

Paper Nr: 13
Title:

Translational Robustness of Neural Networks Trained for Transcription Factor Binding Site Classification

Authors:

Gergely Pap and István Megyeri

Abstract: Classifying DNA sequences based on their protein binding profiles using Deep Learning has enjoyed considerable success in recent years. Although these models can recognize binding sites at high accuracy, their underlying behaviour is unknown. Meanwhile, adversarial attacks against deep learning models have revealed serious issues in the fields of image- and natural language processing related to their black box nature. Analysing the robustness of Transcription Factor Binding Site classifiers urges us to develop adversarial attacks for them. In this work, we introduce shifting as an adversarial data augmentation so that it quantifies the translational robustness. Our results show that despite its simplicity our attack can significantly affect performance. We evaluate two architectures using two data sets with three shifting strategies and train robust models with adversarial data augmentation.
Download

Paper Nr: 14
Title:

Duplicate Detection in a Knowledge Base with PIKA

Authors:

Maxime Prieur, Guillaume Gadek and Bruno Grilheres

Abstract: This paper explores the use of Graph Neural Network models producing node embeddings, in order to solve the not fully addressed problem of detecting similar items stored in a knowledge base. Leveraging pre-trained models for textual semantic similarity, our proposed method PIKA aggregates heterogeneous (structured and unstructured) characteristics of an entity and its neighborhood to produce an embedding vector that can be used in different tasks such as information retrieval or classification tasks. Our method learns specific weights for each information brought by an entity, enabling us to process it in an inductive fashion.
Download

Paper Nr: 18
Title:

Reasoning with Inconsistency-tolerant Fuzzy Description Logics

Authors:

Norihiro Kamide

Abstract: An inconsistency-tolerant fuzzy description logic is introduced and a translation from this logic to a standard fuzzy description logic is constructed. A theorem for embedding the proposed inconsistency-tolerant fuzzy description logic into the standard fuzzy description logic is proven via this translation. A relative decidability theorem for the inconsistency-tolerant fuzzy description logic w.r.t. the standard fuzzy description logic is also proven using this embedding theorem. These proposed logic and translation are intended to effectively handle inconsistent fuzzy knowledge bases. By using the translation, the previously developed algorithms and methods for the standard fuzzy description logic can be re-purposed for appropriately handling inconsistent fuzzy knowledge bases that are described by the proposed logic. Furthermore, an inconsistency-tolerant fuzzy temporal next-time description logic is obtained from the inconsistency-tolerant fuzzy description logic by adding a temporal next-time operator. Similar results as those for the inconsistency-tolerant fuzzy description logic are also obtained for this temporal extension.
Download

Paper Nr: 23
Title:

ScaPMI: Scaling Parameter for Metric Importance

Authors:

Ramisetty Kavya, Jabez Christopher and Subhrakanta Panda

Abstract: Selection of an optimal classifier is an important task in supervised machine learning, and it depends on performance analytics, metric-importance, and domain requirements. This work considers distinct classifiers as decision alternatives and various performance metrics as decision criteria. The weight for each metric is computed by applying an Analytic hierarchy process on the proposed scaling parameter. Multi-criteria decision-making methods consider the performance of classifiers along with metric-weights to generate the ranking order of alternatives. Some typical experimental observations: Random forest is chosen as an optimal classifier by five MCDM methods for liver disorders dataset; Logistic regression, seems optimal for four MCDM methods over hepatitis dataset, and to three methods over heart disease dataset; many such observations discussed in this work may enable developers to choose appropriate classifier for supervised learning systems.
Download

Paper Nr: 24
Title:

SubTST: A Combination of Sub-word Latent Topics and Sentence Transformer for Semantic Similarity Detection

Authors:

Binh Dang, Tran-Thai Dang and Le-Minh Nguyen

Abstract: Topic information has been useful for semantic similarity detection. In this paper, we present a study on a novel and efficient method to incorporate the topic information with Transformer-based models, which is called the Sub-word Latent Topic and Sentence Transformer (SubTST). The proposed model basically inherits the advantages of the SBERT (Reimers and Gurevych, 2019) architecture, and learns latent topics in the sub- word level instead of the document or word levels as previous work. The experimental results illustrate the effectiveness of our proposed method that significantly outperforms the SBERT, and the tBERT (Peinelt et al., 2020), two state-of-the-art methods for semantic textual detection, on most of the benchmark datasets.
Download

Paper Nr: 26
Title:

CoAx: Collaborative Action Dataset for Human Motion Forecasting in an Industrial Workspace

Authors:

Dimitrios Lagamtzis, Fabian Schmidt, Jan Seyler and Thao Dang

Abstract: Human robot collaboration in industrial workspaces where humans perform challenging assembly tasks has become too much; increasingly popular. Now that intention recognition and motion forecasting is being more and more successful in different research fields, we want to transfer that success (and the algorithms making this success possible) to human motion forecasting in an industrial context. Therefore, we present a novel public dataset comprising several industrial assembly tasks, one of which incorporates interaction with a robot. The dataset covers 3 industrial work tasks with robot interaction performed by 6 subjects with 10 repetitions per subject summing up to 1 hour and 58 minutes of video material. We also evaluate the dataset with two baseline methods. One approach is solely velocity-based and the other one is using timeseries classification to infer the future motion of the human worker.
Download

Paper Nr: 33
Title:

An Algorithm for Estimating Answerers’ Performance and Improving Answer Quality Predictions in QA Forums

Authors:

Yonas D. Woldemariam

Abstract: In this study, a multi-components algorithm is developed for estimating answerer performance, largely from a syntactic representation of answer content. The resulting algorithm has been integrated into semantic based answer quality prediction models, and appears to significantly improve all testsets’ baseline results, in the best case scenario. Upto 86% accuracy and 84% F-measure are scored by these models. Also, answer quality classifiers yeild upto 100% recall and 98% precision. Following the transformation of joint syntactic-punctuation information into the identified expertise dimensions (e.g., authoritativeness, analytical, descriptiveness, completeness) that formally define answerer performance, extensive algorithm analyses have been carried on almost 142,246 answers extracted from diverse sets of 13 different QA forums. The analyses prove that incorporating competence information into answer quality models certainly leads to nearly perfect models. Moreover, we found out that the syntactic based algorithm with semantic based models yield better results than answer quality prediction modles built on shallow linguistic or meta-features presented in related works.
Download

Paper Nr: 39
Title:

Traffic Classification of Home Network Devices using Supervised Learning

Authors:

Adriano A. M. De Resende, Pedro A. D. De Melo, Jefferson R. Souza, Renan G. Cattelan and Rodrigo S. Miani

Abstract: Network traffic classification is a relevant tool for computer network management. In the last decade, researchers have been adopting machine learning algorithms to identify different types of traffic in a network. Traffic classification can be used to identify threats and improve the quality of service of networks. Literature in this area usually focuses on using network flows to identify the traffic of specific devices, for example, IoT devices. This paper proposes a network traffic classification model to identify IoT smart home devices and personal computers (PCs). The idea is to evaluate the performance of decision models trained with different devices to identify IoT and non-IoT network traffic. We created two scenarios to mimic the behavior of a home network. In the first scenario, we evaluate how training a model with only PC devices influences the identification of IoT and non-IoT traffic. The second one attempts to assess how well the network traffic of a brand new type of IoT device could be identified using supervised learning. Our results show that the supervised models were able to identify the network traffic; however, their performance varies across the algorithms.
Download

Paper Nr: 41
Title:

P2Ag: Perception Pipeline in Agriculture for Robotic Harvesting of Tomatoes

Authors:

Soofiyan Atar, Simranjeet Singh, Jaison Jose and Kavi Arya

Abstract: Harvesting tomatoes in agriculture is a time-consuming and repetitive task. Different techniques such as accurate detection, classification, and exact location of tomatoes must be utilized to automate harvesting tasks. This paper proposes a perception pipeline (P2Ag) that can effectively harvest tomatoes using instance segmentation, classification, and semantic mapping techniques. P2Ag is highly optimized for embedded hardware in terms of performance, computational power and cost. It provides decision-making approaches for harvesting along with perception techniques, using a semantic map of the environment. This research offers an end- to-end perception solution for autonomous agricultural harvesting. To evaluate our approach, we designed a simulator environment with tomato plants and a stereo-vision sensor. This paper reports results on detecting tomatoes (actual and simulated ) and marking each tomato’s location in 3D space. In addition, the evaluation shows that the proposed P2Ag outperforms the state-of-the-art implementations.
Download

Paper Nr: 43
Title:

A Simple and Effective Convolutional Filter Pruning based on Filter Dissimilarity Analysis

Authors:

F. X. Erick, Shrutika S. Sawant, Stephan Göb, N. Holzer, E. W. Lang and Th. Götz

Abstract: In this paper, a simple and effective filter pruning method is proposed to simplify the deep convolutional neural network (CNN) and accelerate learning. The proposed method selects the important filters and discards the unimportant ones based on filter dissimilarity analysis. The proposed method searches for filters with decent representative ability and less redundancy, discarding the others. The representative ability and redundancy contained in the filter is evaluated by its correlation with currently selected filters and left over unselected filters. Moreover, the proposed method uses an iterative procedure, so that less representative filters can be discarded evenly from the entire model. The experimental analysis confirmed that a simple filter pruning method can reduce floating point operations (FLOPs) of TernausNet by up to 89.65% on an INRIA Aerial Image Labeling dataset with an only marginal drop in the original accuracy. Furthermore, the proposed method shows promising results in comparison with other state-of-the-art methods.
Download

Paper Nr: 46
Title:

Augmenting Reinforcement Learning to Enhance Cooperation in the Iterated Prisoner’s Dilemma

Authors:

Grace Feehan and Shaheen Fatima

Abstract: Reinforcement learning algorithms applied to social dilemmas sometimes struggle with converging to mutual cooperation against like-minded partners, particularly when utilising greedy behavioural selection methods. Recent research has demonstrated how affective cognitive mechanisms, such as mood and emotion, might facilitate increased rates of mutual cooperation when integrated with these algorithms. This research has, thus far, primarily utilised mobile multi-agent frameworks to demonstrate this relationship - where they have also identified interaction structure as a key determinant of the emergence of cooperation. Here, we use a deterministic, static interaction structure to provide deeper insight into how a particular moody reinforcement learner might encourage the evolution of cooperation in the Iterated Prisoner’s Dilemma. In a novel grid environment, we both replicated original test parameters and then varied the distribution of agents and the payoff matrix. We found that behavioural trends from past research were present (with suppressed magnitude), and that the proportion of mutual cooperations was heightened when both the influence of mood and the cooperation index of the payoff matrix chosen increased. Changing the proportion of moody agents in the environment only increased mutual cooperations by virtue of introducing cooperative agents to each other.
Download

Paper Nr: 57
Title:

Deep-Learning-based Fuzzy Symbolic Processing with Agents Capable of Knowledge Communication

Authors:

Hiroshi Honda and Masafumi Hagiwara

Abstract: The authors propose methods for reproducing deep learning models using a symbolic representation from learned deep reinforcement learning models and building agents capable of knowledge communication with humans. It is difficult for humans to understand the behaviour of agents using deep reinforcement learning, and to inform agents of the state of the environment and to receive actions from the agents. In this paper, fuzzified states of the environment and agent actions are represented by rules of first-order predicate logic, and models using symbolic representation are generated by learning such rules. By replacing deep reinforcement learning models with models using a symbolic representation, it is possible for humans to inform the state of the environment and add rules to the agents. As a result of the experiments, the authors can reproduce trained deep reinforcement learning models with high match rate for two types of reinforcement learning simulation environments. Using reproduced models, the authors build agents that can communicate with humans that have yet be realized thus far. This proposed method is the first case of building agents capable of knowledge communication with humans using trained reinforcement learning models.
Download

Paper Nr: 59
Title:

Spike-time Dependent Feature Clustering

Authors:

Zachary S. Hutchinson

Abstract: In this paper, we present an algorithm capable of spatially encoding the relationships between elements of a feature vector. Spike-time dependent feature clustering positions a set of points within a spherical, non- Euclidean space using the timing of spiking neurons. The algorithm uses an Hebbian process to move feature points. Each point is representative of an individual element of the feature vector. Relative angular distances encode relationships within the feature vector of a particular data set. We demonstrate that trained points can inform a feature reduction process. It is capable of clustering features whose relationships extend through time (e.g., spike trains). In this paper, we describe the algorithm and demonstrate it on several real and artificial data sets. This work is the first stage of a larger effort to construct and train artificial dendritic neurons.
Download

Paper Nr: 69
Title:

Prediction of Store Demands by Decision Trees and Recurrent Neural Networks Ensemble with Transfer Learning

Authors:

Nikica Perić, Naomi-Frida Munitić, Ivana Bašljan and Vinko Lešić

Abstract: Simple vehicle routing problem (VRP) algorithms today achieve near-optimal solution and solve problems with a large number of nodes. Recently, these algorithms are upgraded with additional constraints to respect an increasing number of real-world conditions and, further on, adding a predictive character to the optimization. A distinctive contribution lies in taking into account the predictions of orders that are yet to occur. Such problems fall under time series approaches that are most often obtained using statistical methods or historical data heuristics. Machine learning methods have proven to be superior to statistical methods in most of the literature. In this paper, machine learning techniques for predicting the mass of total daily orders for individual stores are further elaborated and tested on historical data of a local retail company. Among the tested methods are Gradient Boosting Decision Tree methods (XGBoost and LightGBM) and methods of Recurrent Neural Networks (LSTM, GRU and their variations using transfer learning). Finally, an ensemble of these methods is performed, which provides the highest prediction accuracy. The final models use the information on historical order quantities and time-related slack variables.
Download

Paper Nr: 70
Title:

Constrained CP-nets Similarity

Authors:

Hassan Alkhiri and Malek Mouhoub

Abstract: The Conditional Preference Network (CP-net) is one of the widely used graphical models for representing and reasoning with qualitative preferences under ceteris paribus (“all else being equal”) assumptions. CP-nets have been extended to Constrained CP-nets (CCP-nets) in order to consider constraints between attributes. Adding constraints will restrict agent preferences, as some of the outcomes become infeasible. Aggregating CCP-nets (representing different agents) can be very relevant for multi-agent and recommender systems. We address this task by defining the notion of similarity between CCP-nets. The similarity is computed using the Hamming distance (between the outcomes of the related pair of CCP-nets) and the number of preference statements shared by both CCP-nets. We propose an algorithm to compute the distance between a pair of CCP-nets, based on the similarity we defined. In order to evaluate the time performance of our proposed algorithm, we conduct several experiments and report the related results.
Download

Paper Nr: 72
Title:

Falsification-aware Semantics for CTL and Its Inconsistency-tolerant Subsystem: Towards Falsification-aware Model Checking

Authors:

Norihiro Kamide and Seidai Kanbe

Abstract: This study introduces two falsification-aware Kripke-style semantics for computation tree logic (CTL). The equivalences among the proposed falsification-aware Kripke-style semantics and the standard Kripke-style semantics for CTL are proven. Furthermore, a new logic, inconsistency-tolerant CTL (ICTL) is semantically defined and obtained from the proposed falsification-aware Kripke-style semantics for CTL by deleting a characteristic condition on the labeling function of the semantics. Because ICTL is regarded as an inconsistency-tolerant and many-valued logic, the proposed semantic framework for CTL and ICTL is regarded as a unified framework for combining and generalizing the standard, inconsistency-tolerant, and many-valued semantic frameworks. This unified semantic framework is useful for generalized model checking, referred to here as falsification-aware model checking.
Download

Paper Nr: 74
Title:

Evolving Evaluation Functions for Collectible Card Game AI

Authors:

Radosław Miernik and Jakub Kowalski

Abstract: In this work, we presented a study regarding two important aspects of evolving feature-based game evaluation functions: the choice of genome representation and the choice of opponent used to test the model. We compared three representations. One simple and limited, based on a vector of weights, and two more complex, based on binary and n-ary trees. On top of this test, we also investigated the influence of fitness defined as a simulation-based function that: plays against a fixed weak opponent, a fixed strong opponent, and the best individual from the previous population. We encoded our experiments in a programming game, Legends of Code and Magic, used in Strategy Card Game AI Competition. However, as the problems stated are of general nature we are convinced that our observations are applicable in the other domains as well.
Download

Paper Nr: 84
Title:

Recommendation System for Student Academic Progress

Authors:

Horea Greblă, Cătălin V. Rusu, Adrian Sterca, Darius Bufnea and Virginia Niculescu

Abstract: The purpose of this work is to study the possible approaches to build a recommendation system that could help students in organizing their work and improving their results. More specifically, we intend to predict grades of a student for future exams, based on his/her previous results and the past grades received by all students from the same series/group. We have tried several machine learning methods for predicting future student grades, and finally we obtained good results, namely a mean absolute prediction error smaller than 1. The best variant proved to be the one based on neural networks that leads to a mean absolute prediction error smaller than 0.5. These results show the practical applicability of our proposed methodology, and consequently, we built, based on these, a practical recommendation system available to students as a web application.
Download

Paper Nr: 85
Title:

Quantifying Student Attention using Convolutional Neural Networks

Authors:

Andreea Coajă and Cătălin V. Rusu

Abstract: In this study we propose a method for quantifying student attention based on Gabor filters, a convolutional neural network and a support vector machine (SVM). The first stage uses a Gabor filter, which extracts intrinsic facial features. The convolutional neural network processes this initial transformation and in the last layer a SVM performs the classification. For this task we have constructed a custom dataset of images. The dataset consists of images from the Karolinska Directed Emotional Faces dataset, from actual high school online classes and from volunteers. Our model showed higher accuracy when compared to other convolutional models such as AlexNet and GoogLeNet.
Download

Paper Nr: 86
Title:

Magic Mirror: I and My Avatar - A Versatile Augmented Reality Installation Controlled by Hand Gestures

Authors:

Alexander K. Seewald and Alexander Pfeiffer

Abstract: Since 2012 we have been building the augmented reality system Magic Mirror based on Kinect V1 and V2’s native API. It relies on the magic mirror metaphor, where a large screen shows a mirrored camera view with overlaid graphical elements. In our case, it shows a different face mesh over the person’s face which reliably tracks face poses in real time while leaving the eyes and mouth of the person visible for interaction and to improve immersion; replaces the background with images that may be changed, smoothly zoomed and dragged; and allows to take screenshots which are automatically printed out on photo cards with an unique QR code linking to its digital twin. Control of the system is primarily via easily learned hand gestures very similar to multitouch screen gestures known from mobile phones and tablets. We have demonstrated the system to the public as well as in private over a wide variety of settings, faces and backgrounds. Here, we explain the challenges inherent in creating high-quality face meshes and textures from 2D images, and how we solved them; describe the different versions of the system, how they differ and their limitations; and demonstrate the usefulness of our system in several applications from people counting and tracking to obtaining height measurements without storing or processing personal data.
Download

Paper Nr: 87
Title:

Empirical Analysis of Limits for Memory Distance in Recurrent Neural Networks

Authors:

Steffen Illium, Thore Schillman, Robert Müller, Thomas Gabor and Claudia Linnhoff-Popien

Abstract: Common to all different kinds of recurrent neural networks (RNNs) is the intention to model relations between data points through time. When there is no immediate relationship between subsequent data points (like when the data points are generated at random, e.g.), we show that RNNs are still able to remember a few data points back into the sequence by memorizing them by heart using standard backpropagation. However, we also show that for classical RNNs, LSTM and GRU networks the distance of data points between recurrent calls that can be reproduced this way is highly limited (compared to even a loose connection between data points) and subject to various constraints imposed by the type and size of the RNN in question. This implies the existence of a hard limit (way below the information-theoretic one) for the distance between related data points within which RNNs are still able to recognize said relation.
Download

Paper Nr: 89
Title:

Towards Robust Continual Learning using an Enhanced Tree-CNN

Authors:

Basile Tousside, Lukas Friedrichsen and Jörg Frochte

Abstract: The ability to perform continual learning and the adaption to new tasks without losing the knowledge already acquired is still a problem that current machine learning models do not address well. This is a drawback, which needs to be tackled for different reasons. On the one hand, conserving knowledge without keeping all of the data over all tasks is a rising challenge with laws like the European General Data Protection Regulation. On the other hand, training models come along with CO2 footprint. In the spirit of a Green AI the reuse of trained models will become more and more important. In this paper we discuss a simple but effective approach based on a Tree-CNN architecture. It allows knowledge transfer from past task when learning a new task, which maintains the model compact despite network expansion. Second, it avoids forgetting, i.e., learning new tasks without forgetting previous tasks. Third, it is cheap to train, to evaluate and requires less memory compared to a single monolithic model. Experimental results on a subset of the ImageNet dataset comparing different continual learning methods are presented.
Download

Paper Nr: 91
Title:

Efficient Removal of Weak Associations in Consensus Clustering

Authors:

N. R. Sinorina, Howard J. Hamilton and Sandra Zilles

Abstract: Consensus clustering methods measure the strength of an association between two data objects based on how often the objects are grouped together by the base clusterings. However, incorporating weak associations in the consensus process can have a negative effect on the quality of the aggregated clustering. This paper presents an efficient automatic approach for removing weak associations during the consensus process. We compare our approach to a brute force method used in an existing consensus function, NegMM, which tends to be rather inefficient in terms of runtime. Our empirical analysis on multiple datasets shows that the proposed approach produces consensus clusterings that are comparable in quality to the ones produced by the original NegMM method, yet at a much lower computational cost.
Download

Paper Nr: 92
Title:

Identifying Problematic Gamblers using Multiclass and Two-stage Binary Neural Network Approaches

Authors:

Kurt D. Buttigieg, Mark A. Caruana and David Suda

Abstract: Responsible gaming has gained traction in recent years due to the harmful nature of compulsive online gambling and the increased awareness on the unfavourable consequences arising from this type of gambling. In Malta, legislation passed in 2018 places the onus of responsibility on online gaming companies has made studying this problem even more important. The focus of this research paper is to apply multistage and two-stage artificial neural networks (ANN), and two-stage Bayesian neural networks (BNN), to the responsible gaming problem by training models that can predict the gambling-risk of a player as a multiclass classification problem. The models are trained using data from gambling session histories provided by a gaming company based in Malta. These models will then be compared using different performance metrics. It is shown that, while all approaches considered have their strengths, multiclass artificial neural networks perform best in terms of overall accuracy while the two-stage Bayesian neural network model performs best in classifying the most important class, the one where the players have a high risk of becoming problematic gamblers, and also second best at classifying the medium risk class.
Download

Paper Nr: 93
Title:

Investigation of Capsule Networks Regarding their Potential of Explainability and Image Rankings

Authors:

Felizia Quetscher, Christof Kaufmann and Jörg Frochte

Abstract: Explainable Artificial Intelligence (AI) is a long-ranged goal, which can be approached from different viewpoints. One way is to simplify the complex AI model into an explainable one, another way uses post- processing to highlight the most important input features for the classification. In this work, we focus on the explanation of image classification using capsule networks with dynamic routing. We train a capsule network on the EMNIST letter dataset and examine the model regarding its explanatory potential. We show that the length of the class specific vectors (squash vectors) of the capsule network can be interpreted as predicted probability and it correlates with the agreement between the decoded image and the original image. We use the predicted probabilities to rank images within one class. By decoding different squash vectors, we visualize the interpretation of the image as the corresponding classes. Eventually, we create a set of modified letters to examine which features contribute to the perception of letters. We conclude that this decoding of squash vectors provides a quantifiable tool towards explainability in AI applications. The explanations are trustworthy through the relation between the capsule network’s prediction and the corresponding visualization.
Download

Paper Nr: 94
Title:

EmBoost: Embedding Boosting to Learn Multilevel Abstract Text Representation for Document Retrieval

Authors:

Tolgahan Cakaloglu, Xiaowei Xu and Roshith Raghavan

Abstract: Learning hierarchical representation has been vital in natural language processing and information retrieval. With recent advances, the importance of learning the context of words has been underscored. In this paper we propose EmBoost i.e. Embedding Boosting of word or document vector representations that have been learned from multiple embedding models. The advantage of this approach is that this higher order word embedding represents documents at multiple levels of abstraction. The performance gain from this approach has been demonstrated by comparing with various existing text embedding strategies on retrieval and semantic similarity tasks using Stanford Question Answering Dataset (SQuAD), and Question Answering by Search And Reading (QUASAR). The multilevel abstract word embedding is consistently superior to existing solo strategies including Glove, FastText, ELMo and BERT-based models. Our study shows that further gains can be made when a deep residual neural model is specifically trained for document retrieval.
Download

Paper Nr: 95
Title:

Social Transmission of Information through Virtual Robotic Agents

Authors:

Owais Hamid, Shruti Chandra, Kerstin Dautenhahn and Chrystopher Nehaniv

Abstract: Social learning includes simple or complex social mechanisms that allow us to understand cooperation and communication in animals, giving them better chances to survive for longer and thrive as a society. Specifically, certain types of social learning such as observational conditioning and stimulus enhancement have been investigated in the context of social information spread between primates. However, not many studies have utilized such social learning mechanisms to study social learning between humans and artificial agents. In the work described here, we seek to understand if and how simple social learning mechanisms can influence human participants using an online game platform with an immersive first person experience built through Unity. Specifically, we designed a study inspired by experiments in behavioural sciences to investigate whether and to what extent, a robotic agent can influence human’s actions. The study compared two conditions in which the robot showed body-language based emotions in a positive or negative manner that could enhance certain stimuli for the human participants and influence their decision making. From this, we wanted to understand whether these effects are socially learned by humans. Objective (position of player in-game) and Subjective (questionnaires) measures were recorded, and markers using the objective data suggest successful social transmission of information. We believe this approach can make a novel contribution to the field of Human Interaction with Artificial Agents.
Download

Paper Nr: 103
Title:

Auxiliary Data Selection in Percolative Learning Method for Improving Neural Network Performance

Authors:

Masayuki Kobayashi, Shinichi Shirakawa and Tomoharu Nagao

Abstract: Neural networks have been evolved significantly at the cost of requiring many input data. However, collecting useful data is expensive for many practical uses, which can be barrier for practical use in real-world applications. In this work, we propose a framework for improving the model performance, in which the model leverages the auxiliary data that is only available during the training. We demonstrate how to (i) train the neural network to perform as though auxiliary data are used during the testing, and (ii) automatically select the auxiliary data during training to encourages the model to generalize well and avoid overfitting to the auxiliary data. We evaluate our method on several datasets, and compare the performance with baseline model. Despite the simplicity of our method, our method makes it possible to get good generalization performance in most cases.
Download

Paper Nr: 105
Title:

Comparing RNN and Transformer Context Representations in the Czech Answer Selection Task

Authors:

Marek Medved, Radoslav Sabol and Aleš Horák

Abstract: Open domain question answering now inevitably builds upon advanced neural models processing large unstructured textual sources serving as a kind of underlying knowledge base. In case of non-mainstream highly- inflected languages, the state-of-the-art approaches lack large training datasets emphasizing the need for other improvement techniques. In this paper, we present detailed evaluation of a new technique employing various context representations in the answer selection task where the best answer sentence from a candidate document is identified as the most relevant to the human entered question. The input data here consists not only of each sentence in isolation but also of its preceding sentence(s) as the context. We compare seven different context representations including direct recurrent network (RNN) embeddings and several BERT-model based sentence embedding vectors. All experiments are evaluated with a new version 3.1 of the Czech question answering benchmark dataset SQAD with possible multiple correct answers as a new feature. The comparison shows that the BERT-based sentence embeddings are able to offer the best context representations reaching the mean average precision results of 83.39% which is a new best score for this dataset.
Download

Paper Nr: 106
Title:

On the Suitability of SHAP Explanations for Refining Classifications

Authors:

Yusuf Arslan, Bertrand Lebichot, Kevin Allix, Lisa Veiber, Clément Lefebvre, Andrey Boytsov, Anne Goujon, Tegawendé Bissyande and Jacques Klein

Abstract: In industrial contexts, when an ML model classifies a sample as positive, it raises an alarm, which is subsequently sent to human analysts for verification. Reducing the number of false alarms upstream in an ML pipeline is paramount to reduce the workload of experts while increasing customers’ trust. Increasingly, SHAP Explanations are leveraged to facilitate manual analysis. Because they have been shown to be useful to human analysts in the detection of false positives, we postulate that SHAP Explanations may provide a means to automate false-positive reduction. To confirm our intuition, we evaluate clustering and rules detection metrics with ground truth labels to understand the utility of SHAP Explanations to discriminate false positives from true positives. We show that SHAP Explanations are indeed relevant in discriminating samples and are a relevant candidate to automate ML tasks and help to detect and reduce false-positive results.
Download

Paper Nr: 108
Title:

Multi-label Classification of Aircraft Heading Changes using Neural Network to Resolve Conflicts

Authors:

Md S. Rahman, Laurent Lapasset and Josiane Mothe

Abstract: An aircraft conflict occurs when two or more aircraft cross at a certain distance at the same time. Aircraft heading changes are the common resolution at the en-route level (high altitude). One or more alternative heading changes are possible to resolve a single conflict. We consider this problem as a multi-label classification problem. We developed a multi-label classification model which provides multiple heading advisories for a given conflict. This model we named CRMLnet is based on the use of a multi-layer neural network that classifies all possible heading resolution in a multi-label classification manner. When compared to other machine learning models that use multiple single-label classifiers such as SVM, K-nearest, and LR, our CRMLnet achieves the best results with an accuracy of 98.72% and ROC of 0.999. The simulated data set which consists of conflict trajectories and heading resolutions we have developed and used in our experiments is delivered to the research community on demand. It is freely accessible online at: https://independent.academia.edu/MDSIDDIQURRAHMAN9.
Download

Paper Nr: 109
Title:

A Singlish Supported Post Recommendation Approach for Social Media

Authors:

Umesha Sandamini, Kusal Rathnakumara, Pasan Pramuditha, Madushani Dissanayake, Disni Sriyaratna, Hansi De Silva and Dharshana Kasthurirathna

Abstract: Social media is an attractive means of communication which people used to exchange information. Post recommendation eliminates the overflooding of information in social media to the users’ news feed by suggesting the best matching information based on users’ preference that in return increase the usability. Social media users use different languages and their variations where most of the Sri Lankan users are accustomed to use Sinhala and Romanized Sinhala. However, post recommendation approaches used in current social media applications do not cater to code-mixed text. Therefore, this paper proposes a novel post recommendation approach that supports Singlish. The study is separated into two major components as language identification and transliteration, and post recommendation. In this study, script identification was performed using regular expressions while a Naïve Bayes classification model that accomplished 97% of accuracy was employed for language identification of Romanized text. Transliteration of Singlish to Sinhala was conducted using a character level seq2seq BLSTM model with a BLEU score of 0.94. Furthermore, Google translation API and YAKE were used for Sinhala-English translation and keyword extraction respectively. Post recommendation model utilized a combination of rule-based and CF techniques that accomplished the RMSE of 0.2971 and MAE of 0.2304.
Download

Paper Nr: 110
Title:

Rethinking Traffic Management with Congestion Pricing and Vehicular Routing for Sustainable and Clean Transport

Authors:

Meghana Kshirsagar, Tanishq More, Rutuja Lahoti, Shreya Adgaonkar, Shruti Jain and Conor Ryan

Abstract: Rapid growth in vehicular congestion increases the challenges of traffic management concerning pollution and infrastructure. Efficient traffic governance can have a significant impact on a country’s economy. To alleviate these challenges, we propose an intelligent integrated traffic management system that manages congestion through cost pricing models to achieve smooth traffic flow. We propose a novel rerouting algorithm and ensemble architecture for vehicle detection and classification, tested on live traffic captured in several Indian cities. The ensemble architectures are designed on a combination of existing pre-trained models. Choice of the ensembles is based on accuracy, model interpretability, and energy efficiency. We show that the second-best ensemble produced operates with significantly less energy and better explainability than our best performer and is still within 3% accuracy of the best performer. Based on predefined road priorities, these ensemble models provide traffic and individual vehicle counts, further fed to our proposed rerouting algorithm as input. The rerouting algorithm then recommends alternative routes and estimated journey time to the user. The paper also presents the results obtained by testing the models on real-time traffic videos from Aurangabad (India) on a GPU/CPU cluster consisting of machines incorporating different GPU hardware.
Download

Paper Nr: 114
Title:

Training Machine Learning Models to Detect Group Differences in Neurophysiological Data using Recurrence Quantification Analysis based Features

Authors:

Gianluca Guglielmo, Travis J. Wiltshire and Max Louwerse

Abstract: Physiological data have shown to be useful in tracking and differentiating cognitive processes in a variety of experimental tasks, such as numerical skills and arithmetic tasks. Numerical skills are critical because they are strong predictors of levels of ability in cognitive domains such as literacy, attention, and understanding contexts of risk and uncertainty. In this work, we examined frontal and parietal electroencephalogram signals recorded from 36 healthy participants performing a mental arithmetic task. From each signal, six RQA-based features (Recurrence Rate, Determinism, Laminarity, Entropy, Maximum Diagonal Line Length and, Average Diagonal Line Length) were extracted and used for classification purposes to discriminate between participants performing proficiently and participants performing poorly. The results showed that the three classifiers implemented provided an accuracy above 0.85 on 5-fold cross-validation, suggesting that such features are effective in detecting performance independently from the specific classifiers used. Compared to other successful methods, RQA-based features have the potential to provide insights into the nature of the physiological dynamics and the patterns that differentiate levels of proficiency in cognitive tasks.
Download

Paper Nr: 120
Title:

Reinforcement Learning Guided by Provable Normative Compliance

Authors:

Emery A. Neufeld

Abstract: Reinforcement learning (RL) has shown promise as a tool for engineering safe, ethical, or legal behaviour in autonomous agents. Its use typically relies on assigning punishments to state-action pairs that constitute unsafe or unethical choices. Despite this assignment being a crucial step in this approach, however, there has been limited discussion on generalizing the process of selecting punishments and deciding where to apply them. In this paper, we adopt an approach that leverages an existing framework – the normative supervisor of (Neufeld et al., 2021) – during training. This normative supervisor is used to dynamically translate states and the applicable normative system into defeasible deontic logic theories, feed these theories to a theorem prover, and use the conclusions derived to decide whether or not to assign a punishment to the agent. We use multi- objective RL (MORL) to balance the ethical objective of avoiding violations with a non-ethical objective; we will demonstrate that our approach works for a multiplicity of MORL techniques, and show that it is effective regardless of the magnitude of the punishment we assign.
Download

Paper Nr: 123
Title:

CogToM-CST: An implementation of the Theory of Mind for the Cognitive Systems Toolkit

Authors:

Fabio Grassiotto, Esther L. Colombini, Alexandre Simões, Ricardo Gudwin and Paula P. Costa

Abstract: This article proposes CogToM-CST, an implementation of a Theory of Mind (ToM) model using the Cognitive Systems Toolkit (CST). Psychological research establishes that ToM deficits are usually associated with mind- blindness, the inability to attribute mental states to others, a typical trait of autism. This cognitive divergence prevents the proper interpretation of other individuals’ intentions and beliefs in a given scenario, typically resulting in social interaction problems. Inspired by the psychological Theory of Mind model proposed by Baron-Cohen, this paper presents a computational implementation exploring the usefulness of the common concepts in Robotics, such as Affordances, Positioning, and Intention Detection, to augment the effectiveness of the proposed architecture. We verify the results by evaluating both a canonical False-Belief task and a subset of tasks from the Facebook bAbI dataset.
Download

Paper Nr: 135
Title:

A Feature Engineering Focused System for Acoustic UAV Payload Detection

Authors:

Yaqin Wang, Facundo E. Fagiani, Kar E. Ho and Eric T. Matson

Abstract: The technology evolution of Unmanned Aerial Vehicles (UAVs) or drones, has made these devices suitable for a wide new range of applications, but it has also raised safety concerns as drones can be used for carrying explosives or weapons with malicious intentions. In this paper, Machine Learning (ML) algorithms are used to identify drones carrying payloads based on the sound signals they emit. We evaluate and propose a feature-based classification. Five individual features, and one combinations of features are used to train four different standard machine learning models: SupportVector Machine (SVM), Gaussian Naive Bayes (GNB), K-Nearest Neighbor (KNN) and a Neural Network (NN) model. The training and testing dataset is composed of sound samples of loaded drones and unloaded drones collected by the team. The results show that the combination of features outperforms the individual ones, with much higher accuracy scores.
Download

Paper Nr: 136
Title:

CycleGAN-based Approach for Masked Face Classification

Authors:

Tomoya Matsubara and Ahmed Moustafa

Abstract: In this paper, we propose a learning model for not only distinguishing whether a person is wearing masks but also classifying the position of the worn masks (mask on my chin, mask on my chin and mouth). First, the synthesized face masks image dataset used for training the model is generated closer to the real world data by CycleGAN. Then, the presence / absence and position of masks are classified using a machine learning model. Experimental results show that this approach provides excellent performance in classifying the presence/ absence and the position of masks.
Download

Paper Nr: 141
Title:

A Survey on Tie Strength Estimation Methods in Online Social Networks

Authors:

Isidoros Perikos and Loizos Michael

Abstract: Social networks constitute an important medium for social interaction where people communicate and formulate relationships in a way similar to what they do in real life. The analysis of the users’ relationships in social networks can lead to new insights into human social behavior. Tie strength constitutes a core aspect of social relationships, which represents the importance of a relationship and the closeness of individuals. Understanding the key features of tie strength in social networks can assist in formulating more efficient user-centric services. This survey paper examines the advances in the area of the analysis of tie strength in social networks. We study the dimensions of tie strength and review the key predictive features for each dimension. We, then, undertake a comparative study of methodologies to model tie strength and examine the key findings. Finally, we discuss open issues and challenges in specifying tie strength.
Download

Paper Nr: 145
Title:

An Implemented System for Cognitive Planning

Authors:

Jorge Fernandez, Dominique Longin, Emiliano Lorini and Frédéric Maris

Abstract: We present a system that implements a framework for cognitive planning. The system allows us to represent and reason about the beliefs, desires and intentions of other agents using an NP-fragment of a multiagent epistemic logic. The system has three components: the belief revision, the planning and the translator modules. They work in an integrated way to firstly capture new information about the world, secondly to plan a sequence of speech acts aimed at achieving a persuasive goal and, finally, to verify satisfiability of the formulas generated at each step of the process. We illustrate how our system can be used to implement a persuasive artificial agent interacting with a human user.
Download

Paper Nr: 151
Title:

Map Matching Algorithm for Large-scale Datasets

Authors:

David Fiedler, Michal Čáp, Jan Nykl and Pavol Žilecký

Abstract: GPS receivers embedded in cell phones and connected vehicles generate series of location measurements that can be used for various analytical purposes. A common preprocessing step of this data is the so-called map matching. The goal of map matching is to infer the trajectory that the device followed in a road network from a potentially sparse series of noisy location measurements. Although accurate and robust map matching algorithms based on probabilistic models exist, they are computationally heavy and thus impractical for processing large datasets. In this paper, we present a scalable map matching algorithm based on Dijkstra’s shortest path method, that is both accurate and applicable to large datasets. Our experiments on a publicly available dataset showed that the proposed method achieves accuracy that is comparable to that of the existing map matching methods using only a fraction of computational resources. As a result, our algorithm can be used to efficiently process large datasets of noisy and potentially sparse location data that would be unexploitable using existing techniques due to their high computational requirements.
Download

Paper Nr: 152
Title:

Micro Junction Agent: A Scalable Multi-agent Reinforcement Learning Method for Traffic Control

Authors:

BumKyu Choi, Jean Seong B. Choe and Jong-kook Kim

Abstract: Traffic congestion is increasing steadily worldwide and many researchers have attempted to employ smart methods to control the traffic. One such approach is the multi-agent reinforcement learning (MARL) scheme wherein each agent corresponds to a moving entity such as vehicles. The aim is to make all mobile objects arrive at their target destination in the least amount of time without collision. However, as the number of vehicles increases, the computational complexity increases, and therefore computation cost increases, and scalability cannot be guaranteed. In this paper, we propose a novel approach using MARL, where the traffic junction becomes the agent. Each traffic junction is composed of four Micro Junction Agents (MJAs) and a MJA becomes the observer and the agent controlling all vehicles within the observation area. Results show that MJA outperforms other MARL techniques on various traffic junction scenarios.
Download

Paper Nr: 153
Title:

Types of Flexible Job Shop Scheduling: A Constraint Programming Experiment

Authors:

Erich C. Teppan

Abstract: The scheduling of jobs is a crucial task in every production company and becomes more and more important in the light of the fourth industrial revolution (Industry 4.0) that aims at fully automated processes. One such problem formulation with big practical relevance is the flexible job shop scheduling problem (FJSSP). Since the classic problem formulation is more general than what can be found in most of nowadays industrial environments, this paper introduces different types of flexibility and investigates how the flexibility type, the amount of allowed flexibility and the presence of machine dependent processing times influence the solution quality that can be achieved by a state-of-the-art constraint solver within limited time. Results show that certain forms of flexibility, higher flexibility factors and the absence of machine dependent processing times can ease the problem.
Download

Paper Nr: 162
Title:

An Asymetric-key Cryptosystem based on Artificial Neural Network

Authors:

Rafael Valencia-Ramos, Luis Zhinin-Vera, Gissela E. Pilliza and Oscar Chang

Abstract: Protect the information has always been important concerns for society, and mainly now in digital era. Currently exists different platforms to manage critical and sensitive information, ranging from bank accounts to social media. All platforms have taken steps to guarantee that the data passing through them is protected from hackers. An essential subject in digital world born, giving place to symmetric and asymmetric key algorithms. Asymmetric key algorithms work by manipulating very big prime numbers, which gives a high level of security but also takes a long time to compute. This paper offers a cryptographic system based on deep learning techniques. The approach avoided the necessity of big prime numbers by using the synaptic weights of an autoencoder neural network as encryption and decryption keys. The suggested method allows for a high amount of unpredictability in the initial and final synaptic weights without compromising the network’s overall performance. The results was shown to be resilient and difficult to break in a theoretical security study with a low computational time.
Download

Paper Nr: 163
Title:

Multi-task Deep Reinforcement Learning for IoT Service Selection

Authors:

Hiroki Matsuoka and Ahmed Moustafa

Abstract: Reinforcement learning has emerged as a powerful paradigm for sequential decision making. By using reinforcement learning, intelligent agents can learn to adapt to the dynamics of uncertain environments. In recent years, several approaches using the RL decision-making paradigm have been proposed for IoT service selection in smart city environments. However, most of these approaches rely only on one criterion to select among the available services. These approaches fail in environments where services need to be selected based on multiple decision-making criteria. The vision of this research is to apply multi-task deep reinforcement learning, specifically (IMPALA architecture), to facilitate multi-criteria IoT service selection in smart city environments. We will also conduct its experiments to evaluate and discuss its performance.
Download

Paper Nr: 164
Title:

Climbing the Ladder: How Agents Reach Counterfactual Thinking

Authors:

Caterina Moruzzi

Abstract: We increasingly rely on automated decision-making systems to search for information and make everyday choices. While concerns regarding bias and fairness in machine learning algorithms have high resonance, less addressed is the equally important question of to what extent we are handing our own role of agents over to artificial information-retrieval systems. This paper aims at drawing attention to this issue by considering what agency in decision-making processes amounts to. The main argument that will be proposed is that a system needs to be capable of reasoning in counterfactual terms in order for it to be attributed agency. To reach this step, automated system necessarily need to develop a stable and modular model of their environment.
Download

Paper Nr: 166
Title:

Ensemble of Patches for COVID-19 X-Ray Image Classification

Authors:

Thiago D. Chen, Gabriel D. Oliveira and Zanoni Dias

Abstract: With the COVID-19 pandemic, several efforts have been made to develop quick and effective diagnoses to assist health professionals in decision-making. In this work, we employed convolutional neural networks to classify chest radiographic images of patients between normal, pneumonia, and COVID-19. We evaluated the division of the images into patches, followed by the ensemble between the specialist networks in each of the image’s parts. As a result, our classifier reached 90.67% in the test, surpassing another method in the literature.
Download

Paper Nr: 171
Title:

Bias Assessment in Medical Imaging Analysis: A Case Study on Retinal OCT Image Classification

Authors:

Gabriel Oliveira, Lucas David, Rafael Padilha, Ana Paula da Silva, Francine de Paula, Lucas Infante, Lucio Jorge, Patricia Xavier and Zanoni Dias

Abstract: Deep learning classifiers can achieve high accuracy in many medical imaging analysis problems. However, when evaluating images from outside the training distribution — e.g., from new patients or generated by different medical equipment — their performance is often hindered, highlighting that they might have learned specific characteristics and biases of the training set and can not generalize to real-world scenarios. In this work, we discuss how Transfer Learning, the standard training technique employed in most visual medical tasks in the literature, coupled with small and poorly collected datasets, can induce the model to capture such biases and data collection artifacts. We use the classification of eye diseases from retinal OCT images as the backdrop for our discussion, evaluating several well-established convolutional neural network architectures for this problem. Our experiments showed that models can achieve high accuracy in this problem, yet when we interpret their decisions and learned features, they often pay attention to regions of the images unrelated to diseases.
Download

Paper Nr: 173
Title:

A Safer Approach to Build Recommendation Systems on Unidentifiable Data

Authors:

Kishor D. Gupta, Akib Sadmanee and Nafiz Sadman

Abstract: In recent years, data security has been one of the biggest concerns, and individuals have grown increasingly worried about the security of their personal information. Personalization typically necessitates the collection of individual data for analysis, exposing customers to privacy concerns. Companies create an illusion of safety to make people feel safe using a mainstream word, “encryption”. Though encryption protects personal data from an external breach, the companies can still exploit personal data collected from users as they own the encryption keys. We present a naive yet secure approach for recommending movies to consumers without collecting any personally identifiable information. Our proposed approach can assist a movie recommendation system understand user preferences using the user’s movie watch-time and watch history only. We conducted a comprehensive and comparative study on the performance of three deep reinforcement learning architectures, namely DQN, DDQN, and D3QN, on the same task. We observed that D3QN outperformed the other two architectures and achieved a precision of 0.880, recall of 0.805, and F1 score of 0.830. The results show that we can build a competitive movie recommendation system using unidentifiable data.
Download

Paper Nr: 175
Title:

A Framework for Generating Playstyles of Game AI with Clustering of Play Logs

Authors:

Yu Iwasaki and Koji Hasebe

Abstract: Many attempts have been made to implement agents for playing games with particular playstyles. Most of these were aimed at generating agents with predetermined playstyles. To this end, they set the reward function to increase the reward as the agent acquires their intended playstyles. However, it is not easy to generate unexpected playstyles through this approach. In this study, we propose a framework to generate multiple playstyles without predefining them. The proposed framework first arranges a set of reward functions regarding the target game and repeats to select a function and make an agent learn with it. Each learned agent is made to play the game, and those whose scores are higher than a predetermined threshold are selected. Finally, each cluster obtained from clustering the play logs (i.e., metrics on the behavior in the game) of the selected agents is considered a playstyle. As a result, it is possible to generate playstyles that play the game well using this procedure. We also applied the proposed framework to a roguelike game, MiniDungeons, and observed that multiple playstyles were generated.
Download

Paper Nr: 178
Title:

An Activation Function with Probabilistic Beltrami Coefficient for Deep Learning

Authors:

Hirokazu Shimauchi

Abstract: We propose an activation function that has a probabilistic Beltrami coefficient for deep neural networks. Activation functions play a crucial role in the performance and training dynamics of deep learning models. In recent years, it has been suggested that the performance of real-valued neural networks can be improved by adding a stochastic perturbation term to the activation function. Meanwhile, numerous studies have been conducted on activation functions of complex-valued neural networks. The proposed approach probabilistically deforms the Beltrami coefficient of complex-valued activation functions. The Beltrami coefficient represents the distortion by mapping at each point. In previous research, when dealing with complex numbers, adding a perturbation term meant applying probabilistic parallel translation from a geometric viewpoint. By contrast, our approach introduces a stochastic perturbation for rotation and scaling. Our experimental results show that the proposed activation function improves the performance of image classification tasks, implying that the suggested activation function produces effective representations during training.
Download

Paper Nr: 179
Title:

Soft Adversarial Training Can Retain Natural Accuracy

Authors:

Abhijith Sharma and Apurva Narayan

Abstract: Adversarial training for neural networks has been in the limelight in recent years. The advancement in neural network architectures over the last decade has led to significant improvement in their performance. It sparked an interest in their deployment for real-time applications. This process initiated the need to understand the vulnerability of these models to adversarial attacks. It is instrumental in designing models that are robust against adversaries. Recent works have proposed novel techniques to counter the adversaries, most often sacrificing natural accuracy. Most suggest training with an adversarial version of the inputs, constantly moving away from the original distribution. The focus of our work is to use abstract certification to extract a subset of inputs for (hence we call it ’soft’) adversarial training. We propose a training framework that can retain natural accuracy without sacrificing robustness in a constrained setting. Our framework specifically targets moderately critical applications which require a reasonable balance between robustness and accuracy. The results testify to the idea of soft adversarial training for the defense against adversarial attacks. At last, we propose the scope of future work for further improvement of this framework.
Download

Paper Nr: 180
Title:

Classification of Table Tennis Strokes in Wearable Device using Deep Learning

Authors:

Nuno M. Ferreira, José M. Torres, Pedro Sobral, Rui Moreira and Christophe Soares

Abstract: Analysis of sports performance using mobile and wearable devices is becoming increasingly popular, helping users improve their sports practice. In this context, the goal of this work has been the development of an Apple Watch application, capable of detecting important strokes in the table tennis sport, using a deep learning (DL) model. A dataset of table tennis strokes has been created based on the watch’s accelerometer and gyroscope sensors. The dataset collection was done in the Portuguese table tennis federation training sites, from several athletes, supervised by their coaches. To obtain the best DL model, three different architecture models where trained, compared and evaluated, using the complete dataset: a LSTM based on Create ML/Core ML frameworks (62.70% F1 score) and two Tensorflow based architectures, a CNN-LSTM (96.02% F1 score) and a ConvLSTM (97.33% F1 score).
Download

Paper Nr: 189
Title:

Fair-Net: A Network Architecture for Reducing Performance Disparity between Identifiable Sub-populations

Authors:

Arghya Datta and S. J. Swamidass

Abstract: In real world datasets, particular groups are under-represented, much rarer than others, and machine learning classifiers will often preform worse on under-represented populations. This problem is aggravated across many domains where datasets are class imbalanced, with a minority class far rarer than the majority class. Naive approaches to handle under-representation and class imbalance include training sub-population specific classifiers that handle class imbalance or training a global classifier that overlooks sub-population disparities and aims to achieve high overall accuracy by handling class imbalance. In this study, we find that these approaches are vulnerable in class imbalanced datasets with minority sub-populations. We introduced Fair-Net, a branched multitask neural network architecture that improves both classification accuracy and probability calibration across identifiable sub-populations in class imbalanced datasets. Fair-Nets is a straightforward extension to the output layer and error function of a network, so can be incorporated in far more complex architectures. Empirical studies with three real world benchmark datasets demonstrate that Fair-Net improves classification and calibration performance, substantially reducing performance disparity between gender and racial sub-populations.
Download

Paper Nr: 190
Title:

Look before You Leap! Designing a Human-centered AI System for Change Risk Assessment

Authors:

Binay Gupta, Anirban Chatterjee, Subhadip Paul, Matha Harika, Lalitdutt Parsai, Kunal Banerjee and Vijay Agneeswaran

Abstract: Reducing the number of failures in a production system is one of the most challenging problems in technology driven industries, such as, the online retail industry. To address this challenge, change management has emerged as a promising sub-field in operations that manages and reviews the changes to be deployed in production in a systematic manner. However, it is practically impossible to manually review a large number of changes on a daily basis and assess the risk associated with these. This warrants the development of an automated system to assess the risk associated with a large number of changes. There are a few commercial solutions available to address this problem but those solutions lack the ability to incorporate domain knowledge and continuous feedback from domain experts into the risk assessment process. As part of this work, we aim to bridge the gap between model-driven risk assessment of change requests and the assessment of domain experts by building a continuous feedback loop into the risk assessment process. Here we present our work to build an end-to-end machine learning system along with the discussion of some of the practical challenges we faced related to extreme skewness in class distribution, concept drift, estimation of the uncertainty associated with the model’s prediction and the overall scalability of the system.
Download

Paper Nr: 191
Title:

Evaluating Two Ways for Mobile Robot Obstacle Avoidance with Stereo Cameras: Stereo View Algorithms and End-to-End Trained Disparity-sensitive Networks

Authors:

Alexander K. Seewald

Abstract: Obstacle avoidance is an essential feature for autonomous robots. Recently, stereo view algorithm challenges have started to focus on fast algorithms with low computational expense, which may enable obstacle avoidance on mobile robots using only stereo cameras. Therefore we have evaluated classical and state-of-the-art stereo salgorithms qualitatively using internal datasets from Seewald (2020), showing that – although improvements are discernable – current algorithms still fail at this task. As it is known (e.g. from Muller et al. (2004)) that deep learning networks trained on stereo views do not rely on view disparity – confirmed by the fact that networks perform almost equally well when trained with only one camera image – we present an alternative network which is end-to-end trained on a simple layer of biologically plausible disparity-sensitive cells and show that it performs equally well as systems trained on raw image data, but must by design rely on view disparity alone.
Download

Paper Nr: 197
Title:

Comparing Monocular Camera Depth Estimation Models for Real-time Applications

Authors:

Abdelrahman Diab, Mohamed Sabry and Amr El Mougy

Abstract: Monocular Depth Estimation (MDE) is a fundamental problem in the field of Computer Vision with ongoing developments. For the case of challenging applications such as autonomous driving, where highly accurate results are required in real-time, traditional approaches fall short due to insufficient information to understand the scene geometry. Novel approaches utilizing deep neural networks show significantly improved results, especially in autonomous driving applications. Nevertheless, there now exists a number of promising approaches in literature and their performance has never been compared head-to-head. In this paper, a detailed evaluation of the performance of four selected deep learning networks is presented. We identify a set of metrics to benchmark the selected approaches from different aspects, especially those related to real-time applications. We analyze the results and present insights into the performance levels of the various approaches.
Download

Paper Nr: 202
Title:

A Logical Characterization of Evaluable Knowledge Bases

Authors:

Alexander Sakharov

Abstract: Evaluable knowledge bases are comprised of non-Horn rules with partial predicates and functions, some of them are defined as recursive functions. This paper investigates logical foundations of the derivation of literals from evaluable knowledge bases without reasoning by contradiction. The semantics of this inference is specified by constrained 3-valued models. The derivation of literals without reasoning by contradiction is characterized by means of sequent calculi with non-logical axioms expressing knowledge base rules and facts. The logical rules of these calculi include only the negation rules, and cut is the only essential structural rule.
Download

Paper Nr: 208
Title:

Towards a Formal Framework for Social Robots with Theory of Mind

Authors:

Filippos Gouidis, Alexandros Vassiliades, Nena Basina and Theodore Patkos

Abstract: A key factor of success for future social robotics entities is going to be their ability to operate in tight collaboration with non-expert human users in open environments. Apart from physical skills, these entities will have to exhibit intelligent behavior, in order both to understand the dynamics of the domain they inhabit and to interpret human intuition and needs. In this paper, we discuss work in progress towards developing a formal framework for endowing intelligent autonomous agents with advanced cognitive skills, central to human-machine interaction, such as Theory of Mind. We argue that this line of work can lay the ground for both theoretical and practical research, and present a number of areas, where such a framework can achieve essential impact for future social and intelligent systems.
Download

Paper Nr: 212
Title:

A Formal Framework for Designing Boundedly Rational Agents

Authors:

Andreas Brännström, Timotheus Kampik, Ramon Ruiz-Dolz and Joaquin Taverner

Abstract: Notions of rationality and bounded rationality play important roles in research on the design and implementation of autonomous agents and multi-agent systems, for example in the context of instilling socially intelligent behavior into computing systems. However, the (formal) connection between artificial intelligence research on the design and implementation of boundedly rational and socially intelligent agents on the one hand and formal economic rationality – i.e., choice with clear and consistent preferences – or instrumental rationality – i.e., the maximization of a performance measure given an agent’s knowledge – on the other hand is weak. In this paper we address this shortcoming by introducing a formal framework for designing boundedly rational agents that systematically relax instrumental rationality, and we propose a system architecture for implementing such agents.
Download

Paper Nr: 214
Title:

Late Bindings in AgentSpeak(L)

Authors:

Frantisek Zboril, Frantisek Vidensky, Radek Koci and Frantisek V. Zboril

Abstract: For agents based on BDI theory, some problems remain open. These include parts of the interpretation of these systems that are nondeterministic in the original specifications, and finding methods for their determinism should lead to improved rationality of agent behaviour. These problems include the choice of a plan suitable for achieving the goal, then the choice of the intention to be pursued by the agent at any given time, and if a language based on predicate logic is used to implement such an agent, then there is also the problem of choosing variable substitutions. One such agent-based system is systems using the AgentSpeak(L) language, which will be the basis for this paper. We will introduce late binding into the interpretation of this language and show that they do not make the agent lose the possibility of achieving the goal by making unnecessary or incorrect substitutions in cases where such a decision is not necessary. We show that with late binding substitutions the agent operates with all possible substitutions given by the chosen plan to the goals in the plan structure, and that these substitutions are always valid with respect to the acts performed so far within this plan.
Download

Paper Nr: 216
Title:

On Equivalence between Linear-chain Conditional Random Fields and Hidden Markov Chains

Authors:

Elie Azeraf, Emmanuel Monfrini and Wojciech Pieczynski

Abstract: Practitioners successfully use hidden Markov chains (HMCs) in different problems for about sixty years. HMCs belong to the family of generative models and they are often compared to discriminative models, like conditional random fields (CRFs). Authors usually consider CRFs as quite different from HMCs, and CRFs are often presented as interesting alternatives to HMCs. In some areas, like natural language processing (NLP), discriminative models have completely supplanted generative models. However, some recent results show that both families of models are not so different, and both of them can lead to identical processing power. In this paper, we compare the simple linear-chain CRFs to the basic HMCs. We show that HMCs are identical to CRFs in that for each CRF we explicitly construct an HMC having the same posterior distribution. Therefore, HMCs and linear-chain CRFs are not different but just differently parametrized models.
Download

Paper Nr: 218
Title:

Learning Optimal Behavior in Environments with Non-stationary Observations

Authors:

Ilio Boone and Gavin Rens

Abstract: In sequential decision-theoretic systems, the dynamics might be Markovian (behavior in the next step is independent of the past, given the present), or non-Markovian (behavior in the next step depends on the past). One approach to represent non-Markovian behaviour has been to employ deterministic finite automata (DFA) with inputs and outputs (e.g. Mealy machines). Moreover, some researchers have proposed frameworks for learning DFA-based models. There are at least two reasons for a system to be non-Markovian: (i) rewards are gained from temporally-dependent tasks, (ii) observations are non-stationary. Rens et al. (2021) tackle learning the applicable DFA for the first case with their ARM algorithm. ARM cannot deal with the second case. Toro Icarte et al. (2019) tackle the problem for the second case with their LRM algorithm. In this paper, we extend ARM to deal with the second case too. The advantage of ARM for learning and acting in non-Markovian systems is that it is based on well-understood formal methods with many available tools.
Download

Paper Nr: 231
Title:

A Machine Learning Approach for Spare Parts Lifetime Estimation

Authors:

Luísa Macedo, Luís Miguel Matos, Paulo Cortez, André Domingues, Guilherme Moreira and André Pilastri

Abstract: Under the Industry 4.0 concept, there is increased usage of data-driven analytics to enhance the production process. In particular, equipment maintenance is a key industrial area that can benefit from using Machine Learning (ML) models. In this paper, we propose a novel Remaining Useful Life (RUL) ML-based spare part prediction that considers maintenance historical records, which are commonly available in several industries and thus more easy to collect when compared with specific equipment measurement data. As a case study, we consider 18,355 RUL records from an automotive multimedia assembly company, where each RUL value is defined as the full amount of units produced within two consecutive corrective maintenance actions. Under regression modeling, two categorical input transforms and eight ML algorithms were explored by considering a realistic rolling window evaluation. The best prediction model, which adopts an Inverse Document Frequency (IDF) data transformation and the Random Forest (RF) algorithm, produced high-quality RUL prediction results under a reasonable computational effort. Moreover, we have executed an eXplainable Artificial Intelligence (XAI) approach, based on the SHapley Additive exPlanations (SHAP) method, over the selected RF model, showing its potential value to extract useful explanatory knowledge for the maintenance domain.
Download

Paper Nr: 237
Title:

A Smart Home Testbed for Evaluating XAI with Non-experts

Authors:

Kevin McAreavey, Kim Bauters and Weiru Liu

Abstract: Smart homes are powered by increasingly advanced AI, yet are controlled by, and affect, non-experts. These non-expert home users are an under-represented stakeholder in the explainable AI (XAI) literature. In this paper we facilitate future XAI research by introducing a family of smart home applications serving as a testbed to evaluate XAI with non-experts. The testbed is a hybrid-AI system spanning several AI disciplines, including machine learning and AI planning. Applications include a smart home battery and smart thermostatic radiator valve (TRV). End-user functionality is representative of leading commercial products and relevant research applications. The testbed is based on a flexible software architecture and web-based user interface, supports a range of AI tools in a modular fashion, and can be easily deployed using inexpensive consumer hardware.
Download

Paper Nr: 245
Title:

Understanding the Scene: Identifying the Proper Sensor Mix in Different Weather Conditions

Authors:

Ziad Elmassik, Mohamed Sabry and Amr El Mougy

Abstract: Autonomous vehicles rely on a variety of sensors for accurate perception and understanding of the scene. Behind these sensors, complex networks and systems perform the driving tasks. Data from the sensors is constantly perturbed by various noise elements, which compromises the reliability of the vehicle’s perception systems. Sensor fusion may be applied to overcome these challenges, especially when the data from the different sensors lead to contradicting results. Nevertheless, weather conditions such as rain, snow, fog, and direct sunlight have an impact on the quality of sensor data, in different ways. This challenge has not been studied in depth, according to the best knowledge of the authors. Accordingly, this paper presents an extensive study of perception systems under different weather conditions, using real-life datasets (nuScenes and the CADCD). We identify a set of evaluation metrics and study the quality of data from different sensors in different scenarios and conditions. Our performance analysis produces insight as to the proper sensor mix that should be used in different weather conditions.
Download

Paper Nr: 246
Title:

Combining YOLO and Deep Reinforcement Learning for Autonomous Driving in Public Roadworks Scenarios

Authors:

Nuno Andrade, Tiago Ribeiro, Joana Coelho, Gil Lopes and A. F. Ribeiro

Abstract: Autonomous driving is emerging as a useful practical application of Artificial Intelligence (AI) algorithms regarding both supervised learning and reinforcement learning methods. AI is a well-known solution for some autonomous driving problems but it is not yet established and fully researched for facing real world problems regarding specific situations human drivers face every day, such as temporary roadworks and temporary signs. This is the core motivation for the proposed framework in this project. YOLOv3-tiny is used for detecting roadworks signs in the path traveled by the vehicle. Deep Deterministic Policy Gradient (DDPG) is used for controlling the behavior of the vehicle when overtaking the working zones. Security and safety of the passengers and the surrounding environment are the main concern taken into account. YOLOv3-tiny achieved an 94.8% mAP and proved to be reliable in real-world applications. DDPG made the vehicle behave with success more than 50% of the episodes when testing, although still needs some improvements to be transported to the real-world for secure and safe driving.
Download

Paper Nr: 247
Title:

Controlled Synthesis of Fibre-reinforced Plastics Images from Segmentation Maps using Generative Adversarial Neural Networks

Authors:

Nicolas Schaaf, Hans A. Zhou, Chrismarie Enslin, Florian Brillowski and Daniel Lütticke

Abstract: The replacement of traditional construction materials with lightweight fibre-reinforced plastics is an accepted way to reduce emissions. By automating quality assurance, errors in production can be detected earlier, avoiding follow-up costs and hard-to-recycle scrap. Deep learning based defect detection systems have shown promising results, but their prediction accuracy often suffers from scarce labelled data in production processes. Especially in the domain of fibre-reinforced plastics, the task remains challenging because of varying textile specific errors. In our work, we applied conditional generative adversarial networks combined with image-to-image translation methods to address data scarcity through generating synthetic images. By training a generative model on image-segmentation pairs, we produce realistic fibre images matching the given segmentation maps. Our model enables control over generated output images of arbitrary fibre shapes and structures, including gaps, ondulations, and folds as error classes. We evaluate our synthetic images based on GAN metrics, feature distribution and show that they are suitable as a data augmentation method to improve the error classification performance of deep neural networks. Thereby, we provide a solution for the manufacturing domain of fibre-reinforced plastics with scarce data, consequently contributing to an automated defect detection system that reduces resource-intensive scrap in the future.
Download

Paper Nr: 249
Title:

YOLOv3: Traffic Signs & Lights Detection and Recognition for Autonomous Driving

Authors:

Rafael Marques, Tiago Ribeiro, Gil Lopes and A. F. Ribeiro

Abstract: Advanced Driver Assistance Systems (ADAS) relates to various in-vehicle systems intended to improve road traffic safety by assisting drivers with improved road awareness, inherent dangers and other drivers nearby. Traffic sign detection and recognition is an integral part of ADAS since these provide information about traffic rules, road conditions, route directions and assistance for safe driving. In addition, traffic sign detection and recognition are essential research topics for safe and efficient driving when considering intelligent transportation systems. An approach to traffic sign/light detection and recognition using YOLOv3 and YOLOv3_tiny is presented in this paper in two different environments. The first is on a simulated and real autonomous driving robot for RoboCup Portuguese Open Autonomous Driving Competition. The robot must detect both traffic signs and lights in real-time and behave accordingly. The second environment is on public roads. A computer vision system inside the car points to the road, detecting and classifying traffic signs/lights (T S/L) in different weather and lighting conditions. YOLOv3 and YOLOv3_tiny were tested on both environments with an extensive hyperparameters search. The final result showcases videos of the two algorithms on the two environments.
Download

Paper Nr: 254
Title:

ALP4AI: Agent-based Learning Platform for Introductory Artificial Intelligence

Authors:

Ramoni O. Lasisi, Connor Philips and Nicholas Hartnett

Abstract: We develop ALP4AI, an Agent-based Learning Platform for Introductory Artificial Intelligence. ALP4AI is a graphical-based tool that is suitable for teaching introductory AI, places emphasis on hands-on learning, and provides for visualization of results. The tool we have developed is suitable for solving problems in the state space search problem domain. It provides for different environments modeling, including, environments that contain obstacles or are obstacle-free, single or multi-agent, and contains single or multi goals. Students can also conduct and report results of experiments using ALP4AI. This project is expected to provide a new frontier of a simple, yet theoretically grounded hands-on learning tool with visualization to aid in AI education and provision of vast resources that benefit the academic community.
Download

Paper Nr: 256
Title:

Experimenting Machine-Learning Algorithms for Morphological Disambiguation of Arabic Texts

Authors:

Bilel Elayeb, Mohamed F. Ettih and Raja Ayed

Abstract: Arabic language is characterized by its complexity and its morphological and orthographic variations including syntactic and semantic diversity of a word. This specificity may cause Arabic morphological ambiguity. We present in this paper a new architecture for morphological disambiguation of Arabic texts. The latter can be treated as a classification problem where the set of morphological features’ values represent classes, and a classification algorithm is used to assign a class to each word’s occurrence based on the context. The first step consists of identifying the correct morphological analysis of a non-vocalized Arabic word using the morphological dependencies extracted from the corpus of vocalized texts. Then, we propose a method of transforming imperfect training datasets into perfect data having precise attributes and certain classes. We experiment this architecture on a set of machine-learning classifiers using a corpus of classic Arabic texts. Results highlight some statistically significant improvement of SVM and Naïve Bayes classifiers in terms of disambiguation rate.
Download

Paper Nr: 258
Title:

A Many-valued Semantics for Multi-agent System

Authors:

Yang Song and Satoshi Tojo

Abstract: We often employ epistemic logic to express the epistemic states of agents. However, it is often too complicated to build a Kripke model because we should consider all possibilities of the knowledge between agents. In this paper, we employ a many-valued logic to express the epistemic states of agents. Thus far, the representations usually show the epistemic state of a single agent, however, we apply the logic to the multi-agent system. Here, we consider that there exist three kinds of epistemic states of known, truth-value unknown, and content unknown. Furthermore, we introduce two kinds of agent communication in our semantics, i.e., teaching and asking, and show how the epistemic states of agents will change.
Download

Paper Nr: 259
Title:

Grid Representation in Neural Networks for Automated Planning

Authors:

Michaela Urbanovská and Antonín Komenda

Abstract: Automated planning and machine learning create a powerful combination of tools which allows us to apply general problem solving techniques to problems that are not modeled using classical planning techniques. In real-world scenarios and complex domains, creating a standardized representation is often a bottleneck as it has to be modeled by a human. That often limits the usage of planning algorithms to real-world problems. The standardized representation is also not a suitable for neural network processing and often requires further transformation. In this work, we focus on presenting three different grid representations that are well suited to model a variety of classical planning problems which can be then processed by neural networks without further modifications. We also analyze classical planning benchmarks in order to find domains that correspond to our proposed representations. Furthermore, we also show that domains that are not explicitly defined on a grid can be represented on a grid with minor modifications that are domain specific. We discuss advantages and drawbacks of our proposed representations, provide examples for many planning benchmarks and also discuss the importance of data and its structure when training neural networks for planning.
Download

Paper Nr: 264
Title:

The Winograd Schema Challenge: Are You Sure That We Are on the Right Track?

Authors:

Nicos Isaak

Abstract: In the past few years, the Winograd Schema Challenge (WSC), the task of resolving ambiguities in carefully- structured sentences, has received considerable interest. According to Levesque, what matters when it comes to the WSC is not a good semblance of intelligent behavior but the behavior itself. In this regard, the WSC has been proposed to understand human behavior as a challenge that could lead to the endowment of machines with commonsense reasoning abilities. Here, we argue that most systems developed so far have typically been designed and evaluated without considering the challenge’s purpose, emphasizing the semblance of intelligence rather than understanding human behavior itself. At the same time, we present an overview of systems developed so far along with a novel developmental-evaluation framework (WSC-Framework 01). The WSC-Framework offers guidelines on what we might need to do to move the field towards the endowment of machines with commonsense reasoning to tackle Winograd schemas the way humans do.
Download

Paper Nr: 266
Title:

Classification Rules Explain Machine Learning

Authors:

Matteo Cristani, Francesco Olvieri, Tewabe C. Workneh, Luca Pasetto and Claudio Tomazzoli

Abstract: We introduce a general model for explainable Artificial Intelligence that identifies an explanation of a Machine Learning method by classification rules. We define a notion of distance between two Machine Learning methods, and provide a method that computes a set of classification rules that, in turn, approximates another black box method to a given extent. We further build upon this method an anytime algorithm that returns the best approximation it can compute within a given interval of time. This anytime method returns the minimum and maximum difference in terms of approximation provided by the algorithm and uses it to determine whether the obtained approximation is acceptable. We then illustrate the results of a few experiments on three different datasets that show certain properties of the approximations that should be considered while modelling such systems. On top of this, we design a methodology for constructing approximations for ML, that we compare to the no-methods approach typically used in current studies on the explainable artificial intelligence topic.
Download

Paper Nr: 267
Title:

Attributed-based Label Propagation Method for Balanced Modularity and Homogeneity Community Detection

Authors:

Jenan Moosa, Wasan Awad and Tatiana Kalganova

Abstract: Community Detection is an expanding field of interest in many scopes, e.g., social science, bibliometrics, marketing and recommendations, biology etc. Various community detection tools and methods have been proposed in the last years. This research is to develop an improved Label Propagation algorithm (Attribute-Based Label Propagation ABLP) that considers the nodes’ attributes to achieve a fair Homogeneity value, while maintaining high Modularity measure. It also formulates an adaptive Homogeneity measure, with penalty and weight modulation, that can be utilized in consonance with the user’s requirements. Based on the literature review, a research gap of employing Homogeneity in Community Detection was identified, and accordingly, Homogeneity as a constraint in Modularity based methods is investigated. In addition, a novel dataset constructed on COVID-19 contact tracing in the Kingdom of Bahrain is proposed, to help identify communities of infected persons and study their attributes’ values. The implementation of proposed algorithm performed high Modularity and Homogeneity measures compared with other algorithms.
Download

Paper Nr: 270
Title:

An Atypical Metaheuristic Approach to Recognize an Optimal Architecture of a Neural Network

Authors:

Abishai Ebenezer M. and Arti Arya

Abstract: The structural design of an Artificial Neural Network (ANN) greatly determines its classification and regression capabilities. Structural design involves both the count of hidden layers and the count of neurons required in each of these hidden layers. Although various optimization algorithms have proven to be good at finding the best topology for a given number of hidden layers for an ANN, there has been little work done in finding both the optimal count of hidden layers and the ideal count of neurons needed in each layer. The novelty of the proposed approach is that a bio-inspired metaheuristic namely, the Water Cycle Algorithm (WCA) is used to effectively search space of local spaces, by using the backpropagation algorithm as the underlying algorithm for parameter optimization, in order to find the optimal architecture of an ANN for a given dataset. Computational experiments have shown that such an implementation not only provides an optimized topology but also shows great accuracy as compared to other advanced algorithms used for the same purpose.
Download

Paper Nr: 272
Title:

Text Classification of English News Articles using Graph Mining Techniques

Authors:

Hasan A. Abdulla and Wasan S. Awad

Abstract: Several techniques can be used in the natural language processing systems to understand text documents, such as, text classification. Text Classification is considered a classical problem with several purposes, varying from automated text classification to sentiment analysis. A graph mining technique for the text classification of English news articles is considered in this research. The proposed model was examined where every text is characterized by a graph that codes relations among the various words. A word's significance to a text is presented by the graph-theoretical degree of a graph's vertices. The proposed weighting scheme can significantly obtain the links between the words that co-appear in a text, producing feature vectors that can enhance the English news articles classification. Experiments have been conducted by implementing the proposed classification algorithms in well-known text datasets. The findings suggest that the proposed text classification using graph mining technique as accurate as other techniques using appropriate parameters.
Download

Paper Nr: 276
Title:

Issue Area Discovery from Legal Opinion Summaries using Neural Text Processing

Authors:

Avi Bleiweiss

Abstract: Applying existed methods of language technology for classifying judicial opinions into their respective issue areas, often requires annotation voting made by human experts. A tedious task nonetheless, further exacerbated by legal descriptions consisting of long text sequences that not necessarily conform to plain English linguistics or grammar patterns. In this paper, we propose instead a succinct representation of an opinion summary joined by case-centered meta-data to form a docket entry. We assembled over a thousand entries from court cases to render our low-resourced target legal domain, and avoided optimistic performance estimates by applying adversarial data split that ensures the most dissimilar train and test sets. Surprisingly, our experimental results show that fine-tuning a pretrained model on standard English recovers issue area prediction by 9 and 8 F1 percentage points over a pretrained model on the legal domain, for macro and weighted average scores, respectively.
Download

Paper Nr: 279
Title:

Development of a New Fundamental Period Formula for Steel Structures Considering the Soil-structure Interaction with the Use of Machine Learning Algorithms

Authors:

Ashley M. van der Westhuizen, George Markou and Nikolaos Bakas

Abstract: The fundamental period of buildings is an important parameter when designing seismic resistant structures. The current formulae proposed in design codes for determining the fundamental period of steel structures cannot accurately predict the fundamental period of real structures. In addition, most of the current formulae only consider the height of the structure in their formulation, while soil structure interaction (SSI) and the orientation of the I-columns that influence the fundamental period are usually neglected. This research focuses on the use of machine learning algorithms to obtain a new formula that accounts for different geometrical features of the superstructure, where the SSI effect is also considered. After training and testing a 40-feature formula, an additional 138 out-of-sample numerical results were used to further test the accuracy of the proposed formula’s prediction abilities. The validation resulted in a correlation of 99.71%, which suggests that the proposed formula exhibits high predictive features for the steel structures considered in this study.
Download

Paper Nr: 282
Title:

Using Machine Learning and Finite Element Modelling to Develop a Formula to Determine the Deflection of Horizontally Curved Steel I-beams

Authors:

Elvis M. Ababu, George Markou and Nikolaos Bakas

Abstract: The use of curved I-beams has been increasing throughout the years as the steel forming industry continues to advance. However, there are often design limitations on such structures due to the lack of recommendations and design code formulae for the estimation of the expected deflection of these structures. This is attributed to the lack of understanding of the behaviour of curved I-beams that exhibit extreme torsion and bending. Thus, currently, there are no formulae readily available for practising engineers to use to estimate the deflection of curved beams. Since the design of light steel structures is often governed by serviceability considerations, this paper aims to analyse the properties of curved steel I-beams and their impact on deflection as well as develop an accurate formula that will be able to predict the expected deflection of these beams. By using a combination of an experimentally validated finite element modelling approach and machine learning. Numerous formulae are developed and tested for the needs of this research work. The final proposed formula, which is the first of its kind, was found to have an average error of 4.11% in estimating the midspan deflection on the test dataset.
Download

Paper Nr: 291
Title:

Training AI to Recognize Realizable Gauss Diagrams: The Same Instances Confound AI and Human Mathematicians

Authors:

Abdullah Khan, Alexei Lisitsa and Alexei Vernitski

Abstract: Recent research in computational topology found sets of counterexamples demonstrating that several recent mathematical articles purporting to describe a mathematical concept of realizable Gauss diagrams contain a mistake. In this study we propose several ways of encoding Gauss diagrams as binary matrices, and train several classical ML models to recognise whether a Gauss diagram is realizable or unrealizable. We test their accuracy in general, on the one hand, and on the counterexamples, on the other hand. Intriguingly, accuracy is good in general and surprisingly bad on the counterexamples. Thus, although human mathematicians and AI perceive Gauss diagrams completely differently, they tend to make the same mistake when describing realizable Gauss diagrams.
Download

Paper Nr: 292
Title:

Snakes in Trees: An Explainable Artificial Intelligence Approach for Automatic Object Detection and Recognition

Authors:

Joanna Isabelle Olszewska

Abstract: Nowadays, the development of smart cities boosts the development of innovative IT technologies based on Artificial Intelligence (AI), such as intelligent agents (IA), which themselves use new algorithms, complex software, and advanced systems. However, due to their expanding number and range of applications as well as their growing autonomy, there is an increased expectation for these intelligent technologies to involve explainable algorithms, dependable software, trustworthy systems, transparent agents, etc. Hence, in this paper, we present a new explainable algorithm which uses snakes within trees to automatically detect and recognize objects. The proposed method involves the recursive computation of snakes (aka parametric active contours), leading to multi-layered snakes where the first layer corresponds to the main object of interest, while the next-layer snakes delineate the different sub-parts of this foreground. Visual features are extracted from the regions segmented by these snakes and are mapped into semantic concepts. Based on these attributes, decision trees are induced, resulting in effective semantic labeling of the objects and the automatic annotation of the scene. Our computer-vision approach shows excellent computational performance on real-world standard database, in context of smart cities.
Download

Paper Nr: 7
Title:

Solving Large Steiner Tree Problems in Graphs for Cost-efficient Fiber-To-The-Home Network Expansion

Authors:

Tobias Müller, Kyrill Schmid, Daniëlle Schuman, Thomas Gabor, Markus Friedrich and Marc Geitz

Abstract: The expansion of Fiber-To-The-Home (FTTH) networks creates high costs due to expensive excavation procedures. Optimizing the planning process and minimizing the cost of the earth excavation work therefore lead to large savings. Mathematically, the FTTH network problem can be described as a minimum Steiner Tree problem. Even though the Steiner Tree problem has already been investigated intensively in the last decades, it might be further optimized with the help of new computing paradigms and emerging approaches. This work studies upcoming technologies, such as Quantum Annealing, Simulated Annealing and nature-inspired methods like Evolutionary Algorithms or slime-mold-based optimization. Additionally, we investigate partitioning and simplifying methods. Evaluated on several real-life problem instances, we could outperform a traditional, widely-used baseline (NetworkX Approximate Solver (Hagberg et al., 2008)) on most of the domains. Prior partitioning of the initial graph and the slime-mold-based approach were especially valuable for a cost-efficient approximation. Quantum Annealing seems promising, but was limited by the number of available qubits.
Download

Paper Nr: 10
Title:

ImpressionNet: A Multi-view Approach to Predict Socio-facial Impressions

Authors:

Rohan K. Gupta and Dakshina R. Kisku

Abstract: The visual facial features do reveal a lot about an individual and can be used to analyse several important social attributes. Existing works have shown that it is possible to learn these attributes through computational models and classify or score subject-faces accordingly. However, we find that there exists local variance in perception. There could be different perspectives of the face which the conventional methods fail to efficiently capture. We also note that Deeper neural networks usually require enough training data and add little to no improvement upon existing ones. In this work, we take social attribute prediction a notch higher and propose a novel multi-view regression approach to incorporate multiple views of face inspired by multi-modal learning. Experimental results show that the proposed approach can achieve superior feature generalisation and diversification on existing datasets using multiple views to improve the coefficient of determination scores and outperforms the state-of-the-art social attribute prediction method. We further propose a method that enables real-time video analysis of multiple subject faces which can have several applications.
Download

Paper Nr: 17
Title:

Negative Selection in Classification using DBLOSUM Matrices as Affinity Function

Authors:

Adil Ibrahim and Nicholas K. Taylor

Abstract: This paper presents a novel affinity function for the Negative Selection based algorithm in binary classification. The proposed method and its classification performance are compared to several classifiers using different datasets. One of the binary classification problems includes medical testing to determine if a patient has a particular disease or not. The DBLOSUM in Negative Selection classifier appears to be best suited to classification tasks where false negatives pose a major risk, such as in medical screening and diagnosis. It is more likely than most techniques to result in false positives, but it is as accurate, if not more accurate than most other techniques.
Download

Paper Nr: 19
Title:

A Hybrid Model based on Convolutional Neural Networks and Long Short-term Memory for Rest Tremor Classification

Authors:

Jihen Fourati, Mohamed Othmani and Hela Ltifi

Abstract: Parkinson’s disease is a neurodegenerative disease, in which tremor is the main symptom. Deep brain stimulation can help manage a broad range of neurological ailments such as Parkinson’s disease. It involves electrical impulses delivered to specific targets in the brain, with the purpose of altering or modulating neural functioning. Security is playing a vital role in protecting healthcare gadgets from unauthorized access or modification. Our purpose is to adopt deep learning methodologies to classify resting tremors. To achieve this purpose, a novel approach for resting tremor classification in patients with Parkinson’s disease using a hybrid model based on convolutional neural networks and long short-term memory is proposed. This research exploits the high-level feature extraction of the convolutional neural network model and the potential capacity to capture long-term dependencies of the long short-term memory model. The performed experiments demonstrate that our proposed approach outperforms the best result for other state-of-the-art methods.
Download

Paper Nr: 42
Title:

LCPP: Low Computational Processing Pipeline for Delivery Robots

Authors:

Soofiyan Atar, Simranjeet Singh, Srijan Agrawal, Ravikumar Chaurasia, Shreyas Sule, Sravya Gadamsetty, Aditya Panwar, Amit Kumar and Kavi Arya

Abstract: Perception techniques in novel times have enormously improved in autonomously and accurately predicting the ultimate states of the delivery robots. The precision and accuracy in recent research lead to high computation costs for autonomous locomotion and expensive sensors and server dependency. Low computational algorithms for delivery robots are more viable as compared to pipelines used in autonomous vehicles or prevailing delivery robots. A blend of different autonomy approaches, including semantic segmentation, obstacle detection, obstacle tracking, and high fidelity maps, is presented in our work. Moreover, this method comprises low computational algorithms feasible on embedded devices with algorithms running more efficiently and accurately. Research also analyzes state-of-the-art algorithms via practical applications. Low computational algorithms have a downside of accuracy, which is not as proportional as computation. Finally, the research proposes that this algorithm will be more realizable as compared to Level 5 autonomy for delivery robots.
Download

Paper Nr: 53
Title:

Exploring Contextualized Tag-based Embeddings for Neural Collaborative Filtering

Authors:

Tahar-Rafik Boudiba and Taoufiq Dkaki

Abstract: Neural collaborative filtering approaches are mainly based on learning user-item interactions. Since in collaborative systems, there are several contents surrounding users and items, essentially user reviews or user tags these personal contents are valuable information that can be leveraged with collaborative filtering approaches. In this context, we address the problem of integrating such content into a neural collaborative filtering model for rating prediction. Such content often represented using the bag of words paradigm is subject to ambiguity. Recent approaches suggest the use of deep neuronal architectures as they attempt to learn semantic and contextual word representations. In this paper, we extended several neural collaborative filtering models for rating prediction that were initially intended to learn user-item interaction by adding textual content. We describe an empirical study that evaluates the impact of using static or contextualized word embeddings with a neural collaborative filtering strategy. The presented models use dense tag-based user and item representations extracted from pre-trained static Word2vec and contextual BERT. The Models were adapted using MLP and Autoencoder architecture and evaluated on several MovieLens datasets. The results showed good improvements when integrating contextual tag embeddings into such neural collaborative filtering architectures.
Download

Paper Nr: 55
Title:

Using ConvNet for Classification Task in Parallel Coordinates Visualization of Topologically Arranged Attribute Values

Authors:

Piotr Artiemjew and Sławomir K. Tadeja

Abstract: In this work, we assess the classification capability of visualized multidimensional data used in the decision- making process. We want to investigate if classification carried out over a graphical representation of the tabular data allows for statistically greater efficiency than the dummy classifier method. To achieve this, we have used a convolutional neural network (ConvNet) as the base classifier. As an input into this model, we used data presented in the form of 2D curves resulting from the Parallel Coordinates Plot (PCP) visualization. Our initial results show that the topological arrangement of attributes, i.e., the shape formed by the PCP curves of individual data items, can serve as an effective classifier. Tests performed on three different real-world datasets from the UCI Machine Learning Repository confirmed that classification efficiency is significantly higher than in the case of dummy classification. The new method provides an interesting approach to the classification of tabular data and offers a unique perspective on classification possibilities. In addition, we examined relevant information content potentially helpful in building hybrid classification models, e.g., in the classifier committee model. Moreover, our method can serve as an enhancement of the PCP visualization itself. Here, we can use our classification technique as a form of double-checking for the pattern identification task performed over PCP by the users.
Download

Paper Nr: 58
Title:

Object-less Vision-language Model on Visual Question Classification for Blind People

Authors:

Tung Le, Khoa Pho, Thong Bui, Huy Tien Nguyen and Minh Le Nguyen

Abstract: Despite the long-standing appearance of question types in the Visual Question Answering dataset, Visual Question Classification does not received enough public interest in research. Different from general text classification, a visual question requires an understanding of visual and textual features simultaneously. Together with the enthusiasm and novelty of Visual Question Classification, the most important and practical goal we concentrate on is to deal with the weakness of Object Detection on object-less images. We thus propose an Object-less Visual Question Classification model, OL–LXMERT, to generate virtual objects replacing the dependence of Object Detection in previous Vision-Language systems. Our architecture is effective and powerful enough to digest local and global features of images in understanding the relationship between multiple modalities. Through our experiments in our modified VizWiz-VQC 2020 dataset of blind people, our Object-less LXMERT achieves promising results in the brand-new multi-modal task. Furthermore, the detailed ablation studies show the strength and potential of our model in comparison to competitive approaches.
Download

Paper Nr: 63
Title:

A Constraint Programming Model for the Scheduling Problem with Flexible Maintenance under Human Resource Constraints

Authors:

Meriem Touat, Belaid Benhamou and Fatima B. Tayeb

Abstract: In this work, we tackle the scheduling problem that considers both production and flexible preventive maintenance on a single machine where the human resource constraints (the availability and the competence) are taken into account. We propose a mathematical formulation for the problem that is expressed in the constraint programming (CP) paradigm as a set of constraints. This CP modeling had been implemented using Ilog CP Optimizer. Experiments were first carried out on small instances to compare our CP implementation with that one carried out in Mixed Integer Linear Program programming (MILP) presented in (Touat et al., 2021), then the CP implementation had been tested on large instances and encouraging results were obtained.
Download

Paper Nr: 64
Title:

A Lexicon-based Collaborative Filtering Approach for Recommendation Systems

Authors:

Mara Deac-Petruşel

Abstract: Users purchasing items from e-commerce websites are expressing their satisfaction and sentiment about their acquisition using text-based reviews and numerical ratings. Traditional collaborative filtering techniques are entirely dependent on the users’ scalar ratings, which are lacking any semantic explanation of the users’ preferences. This approach was designed to explore the text-based item evaluation using a Sentiment Analysis Lexicon. The proposed lexicon-based k nearest neighbors collaborative filtering technique replaces the numerical rating with a computed sentiment rating in the neighborhood determination step. The conducted experiments reveal that the resulting text-based recommendation system produces reliable values in terms of mean absolute error and root mean square error and accurate recommendations for users.
Download

Paper Nr: 67
Title:

Whale Optimization-based Prediction for Medical Diagnostic

Authors:

Ali R. Hosseinabadi, Mehdi Sadeghilalimi, Morteza B. Shareh, Malek Mouhoub and Samira Sadaoui

Abstract: This study aims to improve disease detection accuracy by incorporating a discrete version of the Whale Optimization Algorithm (WOA) into a supervised classification framework (KNN). We devise the discrete WOA by redefining the related components to operate on discrete spaces. More precisely, we redefine the notion of distance (between individuals in WOA), and propose a random exploration function to include more diversity. The latter includes the random move defined in the WOA algorithm, as well as two other random techniques based on the crossover and mutation operators. To assess the performance of our proposed method, we conducted experiments on two benchmark medical datasets. The results demonstrate the efficacy of the hybrid approach, WOA+KNN.
Download

Paper Nr: 71
Title:

Discrete Mother Tree Optimization and Swarm Intelligence for Constraint Satisfaction Problems

Authors:

Wael Korani and Malek Mouhoub

Abstract: The Constraint Satisfaction Problem (CSP) is a powerful framework for a wide variety of combinatorial problems. The CSP is known to be NP-complete, and many algorithms have been developed to tackle this challenge in practice. These algorithms include the backtracking technique, improved with constraint propagation and variable ordering heuristics. Despite its success, backtracking still suffers from its exponential time cost, especially for large to solve problems. Metaheuristics, including local search and nature-inspired methods, can be an alternative that trades running time for the quality of the solution. Indeed, these techniques do not guarantee to return a complete solution, nor can they prove the inconsistency of the problem. They are, however time-efficient, thanks to their polynomial running time. In particular, nature-inspired techniques can be very effective if designed with a good exploitation/exploration balance during the search. To solve CSPs, we propose two discrete variants of two known nature-inspired algorithms. The first one is an adaptation of the Mother Tree Optimization (MTO). In contrast, the second is an extension of the Particle Swarm Optimization (PSO) with a new operator that we propose. Both variants rely on a heuristic that gathers information about constraints violations during the search. The latter will then be used to update candidate solutions, following a given topology for MTO, and position/velocity equations for PSO. To assess the performance of both methods, we conducted several comparative experiments, considering other known systematic methods and metaheuristics. The results demonstrate the effectiveness of both methods.
Download

Paper Nr: 75
Title:

Analysing the Sentiments in Online Reviews with Special Focus on Automobile Market

Authors:

Ayman Yafoz, Farial Syed, Malek Mouhoub and Lisa Fan

Abstract: Analysing the sentiments in online reviews assists in understanding customers’ satisfaction with a provided service or product, which gives the industry an opportunity to enhance the quality of their commodity, increase sales volume, develop marketing strategies, improve response to customers, promote customer satisfaction, and enhance the industry image. However, the studies focusing on applying machine learning algorithms and word embedding models, as well as deep learning techniques to classify the sentiments in reviews extracted from automobile forums, are arguably limited, and to fill this gap, this research addressed this area. Moreover, the research concentrated on categorizing positive, negative, and mixed sentiment categories in online forum reviews. The procedures for gathering and preparing the dataset are illustrated in this research. To perform the classification task, a set of models which include supervised machine learning, deep learning, and BERT word embedding is adopted in this research. The results show that the combination of the BERT word embedding model with the LSTM model produced the highest F1 score. Finally, the paper lays out recommendations to enhance the proposed system in future studies.
Download

Paper Nr: 77
Title:

Incremental Feature Learning for Fraud Data Stream

Authors:

Armin Sadreddin and Samira Sadaoui

Abstract: Our research addresses the actual behavior of the credit-card fraud detection environment where financial transactions containing sensitive data must not be amassed in a considerable amount to train robust classifiers. We introduce an adaptive learning approach that adjusts frequently and efficiently to new transaction chunks; each chunk is discarded after each training step. Our approach combines transfer learning and incremental feature learning. The former improves the feature relevancy for subsequent chunks, and the latter increases performance during training by dynamically determining the optimal network architecture for each new chunk. We show the effectiveness and efficiency of our approach experimentally on an actual fraud dataset.
Download

Paper Nr: 82
Title:

Dynamically Generated Question Answering Evidence using Efficient Context-preserving Subdivision

Authors:

Avi Bleiweiss

Abstract: Recently published datasets for open-domain question answering follow question elicitation from a fairly small snippet of Wikipedia content. Often centered around an article section, the evidence is further subdivided into context-unaware passages of uniform token-lengths to found the basic retrieval units. In this study we hypothesized that splitting a section perceived as an opaque text fragment may hinder quality of answer span predictions. We propose to dynamically draw content corresponding to an article-section url from the most updated online Wikipedia rather than from an archived snapshot. Hence approaching space complexity of O(1), downward from O(n) for a dataset that is fully populated with static context. We then parse the url bound content and feed our neural retriever with a list of paragraph-like html elements that preserve context boundaries naturally. Using knowledge distillation from a sustainable language model pretrained on the large SQuAD 2.0 dataset to the state-of-the-art QuAC domain, shows that during inference our natural context split recovered answer span predictions by 7.5 F1 and 4.1 EM points over a synthetic distribution of fixed-length passages.
Download

Paper Nr: 99
Title:

Classification of Video Viewing Task Types and Recommendation of Videos

Authors:

Tatsuro Ide and Hiroshi Hosobe

Abstract: YouTube is one of the largest and most sophisticated recommendation systems and a useful source of information for users. In video search on YouTube, even the same user may have different purposes in mind depending on the user’s state. However, videos are recommended based on the relevance of videos and the user's viewing history, regardless of the user's state. This paper proposes a classification of video viewing task types based on the user’s behavioral characteristics. By classifying the user's purpose as a task type, it enables higher-order recommendation that fits the task type. Behavioral characteristics are momentary characteristics of the user that appear from actions such as screen scrolling. The system implicitly records the user’s actions, classifies the task type based on these parameters, and recommends the related video list on a mobile application that imitates YouTube. We conducted experiments to evaluate classification of task types and recommendation of videos.
Download

Paper Nr: 118
Title:

Water Consumption Demand Pattern Analysis using Uncertain Smart Water Meter Data

Authors:

Milad Khaki and Nasim Mortazavi

Abstract: Wireless ‘smart’ water meters that allow functionalities such as demand response, leak alerts, identification of characteristic demand patterns, and detailed consumption analysis are becoming an essential part of water infrastructure in many countries. To achieve these benefits, the meter data needs to be error-free, which is not necessarily available in practice due to ‘dirtiness’ or ‘uncertainty’ of data, which is mostly unavoidable. Additionally, by analyzing the smart meter data and finding demand patterns, it is possible to provide insights to the municipalities to improve their distribution network, better understand demand characteristics, identify the consumers that are the main sources of shaping the high consumption peaks. This paper investigates solutions to mine the uncertain data, ensures the validity of results, and evaluates the impact of dirty data on data analysis results. Once the reliability of results is ensured, the evaluation results can be used for informed decision-making on water planning strategies. Secondly, the consumption pattern of a city equipped with 25 thousand water consumers is analyzed, and weekly consumption profiles over an entire year are presented for single-family residential consumers. Additionally, a systematic study of the errors existing in large-scale smart water meter deployments is performed to better understand the nature of errors in such data sources, particularly at the first stages of implementation of smart metering infrastructure. Also, the sensitivity of the results to various types of errors in a big data system is presented and investigated.
Download

Paper Nr: 121
Title:

Upper Confident Bound Fuzzy Q-learning and Its Application to a Video Game

Authors:

Takahiro Morita and Hiroshi Hosobe

Abstract: This paper proposes upper confident bound (UCB) fuzzy Q-learning by combining fuzzy Q-learning and the UCBQ algorithm and applies it to a video game. The UCBQ algorithm improved the action selection method called the UCB algorithm by applying it to Q-learning. The UCB algorithm selects the action with the highest UCB value instead of a value estimate. Since the UCB algorithm is based on the premise that any unselected actions are selected and value estimates are obtained, the number of unselected actions becomes small, and it is able to prevent local optimal solutions. The proposed method aims to promote the efficiency of learning by reducing unselected actions and preventing the Q value from becoming a local optimal solution in fuzzy Q-learning. This paper applies the proposed method to a video game called Ms. PacMan and presents the result of an experiment on finding optimum values in the method. Its evaluation is conducted by comparing the game scores with the scores obtained by a previous fuzzy Q-learning method. The result shows that the proposed method significantly reduced unselected actions.
Download

Paper Nr: 158
Title:

Logical Structure-based Pretrained Models for Legal Text Processing

Authors:

Ha Thanh Nguyen and Le Minh Nguyen

Abstract: In recent years, we have witnessed breakthroughs in natural language processing coming from pretrained models based on the Transformer architecture. In the field of legal text processing, a special sub-domain of NLP, pretrained models also show promising results. For a legal sentence, although the natural language is used for expression, the real meaning lies in its logical structure. From that observation, we have a hypothesis that the knowledge of recognizing logical structures can support deep learning models to understand the legal text better and achieve a higher performance in the related tasks. To verify our assumption, we design a novel framework to inject the knowledge about recognizing the requisite and effectuation part of a law sentence into Transformer models. Our proposed method is effective and general. By our experiments, we provide informative results about our approach and its performance compared with the baselines.
Download

Paper Nr: 159
Title:

The Impact of COVID-19 on Crime: A Study from the Spatial-temporal Perspective in the Montgomery County, AL

Authors:

Long Ma and Connor H. Wu

Abstract: The policies curbing the spread of COVID-19 can influence the chance of committing a crime. This study aimed to investigate the impacts of COVID-19 on the spatial and temporal patterns of crime in Montgomery City, AL, by wavelet analysis, spatial point test, and machine learning tools. We obtained the crime case records between January 1, 2015 to March 12, 2021 from the police department in the City of Montgomery, and we downloaded demographical data from the U.S. Census. Results show that the overall crime rate in Montgomery decreased during the COVID-19 pandemic. However, crime rates would increase in a shorter time than COVID-19 confirmed cases when the social activities increased. Meanwhile, spatial distributions of simple assault, burglary, and vehicle theft had clustered in Montgomery business and shopping areas. These findings are helpful for the police institution in preventing and minimizing crimes as new COVID-19 variants emerge in the future.
Download

Paper Nr: 167
Title:

Leveraging Event Marketing Performance using AI in Facial Recognition

Authors:

Peter Khallouf and Christine Markarian

Abstract: With the advances in technology and the rapid changes in human-technology interactions, the event marketing field has seen major developments over the past years. Despite its remarkable growth, many aspects of event marketing do not yet align with the best available technologies. In this paper, we aim to leverage event marketing performance using artificial intelligence techniques. We design a framework that optimizes attendee-feedback generation using a facial-recognition algorithm. The framework measures attendees’ engagement levels by periodically extracting attendee facial features during a session, categorizing them into seven states of emotions (anger, disgust, fear, happiness, neutral, sadness, and surprise), and then analyzing session engagements based on the obtained results. These measurements are then used to give insights about an event’s performance during and after sessions, thus improving the overall performance of a given event. The proposed framework is easy-to-implement, time-efficient, and cost-effective.
Download

Paper Nr: 172
Title:

Barriers to the Practical Adoption of Federated Machine Learning in Cross-company Collaborations

Authors:

Tobias Müller, Nadine Gärtner, Nemrude Verzano and Florian Matthes

Abstract: Research in federated machine learning and privacy-enhancing technologies has spiked recently. These technologies could enable cross-company collaboration, which yields the potential of overcoming the persistent bottleneck of insufficient training data. Despite vast research efforts and potentially large benefits, these technologies are only applied rarely in practice and for specific use cases within a single company. Among other things, this little and specific utilization can be attributed to a small amount of libraries for a rich variety of privacy-enhancing methods, cumbersome design of end-to-end privacy-enhancing pipelines and unwieldy cus- tomizability to needed requirements. Hence, we identify the need for an easy-to-use privacy-enhancing tool to support and enable cross-company machine learning, suitable for varying scenarios and easily adjustable to the desired corresponding privacy-utility desiderata. This position paper presents the starting point for our future work aiming at the development of the described application.
Download

Paper Nr: 174
Title:

Detection of Emotion Categories’ Change in Speeches

Authors:

Anwer Slimi, Henri Nicolas and Mounir Zrigui

Abstract: In the past few years, a lot of research has been conducted to predict emotions from speech. The majority of the studies aim to recognize emotions from pre-segmented data with one global label (category). Despite the fact that emotional states are constantly changing and evolving across time, the emotion change has gotten less attention. Mainly, the exiting studies focus either on predicting arousal-valence values or on detecting the instant of the emotion change. To the best of the authors knowledge, this is the first paper that addresses the emotion category change (i.e., predicts the classes existing in a signal such as angry, happy, sad etc.). As a result of that, we propose a model based on the Connectionist Temporal Classification (CTC) loss, along with new evaluation metrics.
Download

Paper Nr: 186
Title:

Towards a Knowledge Graph-specific Definition of Digital Transformation: An Account Networking View for Auditing

Authors:

Florina L. Covaci, Robert A. Buchmann and Radu Dragos

Abstract: This paper reports on an experimental Digital Transformation project where RDF graphs are adopted in an organization’s accounting and document management system as a novel approach to accounting digitization, going beyond traditional ERP systems to enable account-centric network analysis and more insightful master data management - having accounting contextualized in a relationship-rich Knowledge Graph that captures some of the tacit knowledge that accountants and auditors apply during their common tasks. Legacy ERP systems that are based on relational databases face challenges when aggregating information regarding the transactions that an account was involved in, which sometimes involve multihop JOINs, links to contextual documents that may reside elsewhere, or rules that mimic (at least partially) an auditor’s reasoning. The paper reports on a knowledge capture effort for mapping accounting information into an RDF graph in order to overcome limitations of legacy systems with auditing support, currently implemented in a feasibility demonstrator of low Technological Readiness Level. As theoretical implications, we also derive from this experience a novel, specialized definition of Digital Transformation.
Download

Paper Nr: 209
Title:

How to Simplify Law Automatically? A Study on South Korean Legislation and Its Simplified Version

Authors:

Stefanie Urchs, Akshaya Muralidharan and Florian Matthes

Abstract: People with a low literacy level have problems understanding complex texts. Especially legal texts can be challenging. Automatic Text Simplification (TS) can help to make the legal text more accessible. However, most TS research is based on Wikipedia articles and newspaper articles. To be able to use automatic TS on the legal text we have to understand what constitutes simple legal text. Therefore, we examine the English translation of South Korean legislation and its official simplification. Subsequently, we use state of the art TS models on the legislation text. The models simplify the text only quantitatively lacking in retaining the context of the original text.
Download

Paper Nr: 219
Title:

Predicting Trains Delays using a Two-level Machine Learning Approach

Authors:

Hassiba Laifa, Raoudha Khcherif and Henda Ben Ghezala

Abstract: Train delay is a critical problem in railway systems. A previous prediction of delays is a critical issue advantageous for passengers to re-plan their journeys more reliably. It is also essential for railway operators to control the feasibility of timetable realization for more efficient train schedules. This paper aims to present a novel two-level Light Gradient Boosting Machine (LightGBM) approach that combines classification and regression in a hybrid model. It was proposed to predict passenger train delays on the Tunisian railway. The first level indicates the class of delay, where the delays are divided into intervals of 5 minutes ([0,5], [6,10], …, [>60]), 13 classes in total were obtained. The second level then predicts the actual delay in minutes, considering the expected delay class at the first level. This model was trained and tested based on the historical data of train operation collected by the Tunisian National Railways Company (SNCFT) and infrastructure characteristics. Our methodology consists of the following phases: data collection, data cleaning, complete data analysis, feature engineering, modeling and evaluation. The obtained results indicate that the two-level approach based on the LightGBM model outperforms the one-level method. It also outperformed the benchmark models.
Download

Paper Nr: 222
Title:

Usage of Stacked Long Short-Term Memory for Recognition of 3D Analytic Geometry Elements

Authors:

Anca-Elena Iordan

Abstract: For accomplish automatic solving, the capacity to comprehend problems of 3D analytic geometry formulated in natural language is a laborious and stimulating open research theme. For this reason, this research work attempts the achievement of a parser compounded of two important parts: the parsing module and the learning module. The accomplishment of the parsing module requires the design of a method for engendering the series of actions required to acquire the UCCA graph corresponding with a phrase from a 3D analytic geometry problem. In order to design the learning module, is used a recurrent neural network of the Stacked Long Short-Term Memory category, thereby being realized an automatic parsing system. To achieve this goal, the proposed novel solution is accomplished through the usage of Python programming language.
Download

Paper Nr: 223
Title:

BRL: A Toolkit for Learning How an Agent Performs Belief Revision

Authors:

Aaron Hunter and Konstantin Boyarinov

Abstract: Belief revision occurs when an agent receives new information that may conflict with their current beliefs. This process can be modelled by a formal belief revision operator. However, in a practical scenario, simply defining abstract revision operators is not sufficient. A truly intelligent agent must be able to observe how others have revised their beliefs in the past, and use this information to predict how they will revise their beliefs in the future. In other words, an agent must be able to learn the mental model that is used by other agents. This process involves combining two traditionally distinct areas of Artificial Intelligence to produce a general reasoning system. In this paper, we discuss challenges faced in using various learning approaches to learn belief revision operators. We then present the BRL toolkit: software can learn the revision operator an agent is using based on past revisions. This is a tool that bridges formal reasoning and machine learning to address a common problem in practical reasoning. Accuracy and efficiency of the approach are discussed.
Download

Paper Nr: 224
Title:

Vaccination Planning in Peru using Constraint Programming

Authors:

Willy Ugarte

Abstract: Vaccination has been proven to be the most effective method to prevent infectious diseases, specially nowadays with the global pandemic of CoViD19. Millions of people are not immunized yet in various countries because of low vaccine availability resulting from inefficiencies and/or lack of access to the vaccines. We propose a constraint programming model, kwnown as Constraint Satisfaction Problem (CSP) as a distribution model for vaccination to address the unique characteristics and challenges facing vaccine dose assignation. This CSP model capture the uncertainties of demand for vaccinations such as the age range of the vaccination campaign and the location of vaccination centers. The objective is to maximize the percentage of fully immunized people facilitating the access by location and capacity of the vaccination centers while respecting the health ministry dispositions (e.g., age range, number of doses, etc.). Our research examines how these can be optimized with a constraint optimization problem in a single objective function. We tested the model using Peru open data on vaccination planning of their national health ministry. We make many experiments to show the feasibility of our proposal to increase their immunization coverage.
Download

Paper Nr: 248
Title:

Disruption Management of ASAE’s Inspection Routes

Authors:

Miguel M. Ferreira, Henrique L. Cardoso, Luís P. Reis, Telmo Barros and João P. Machado

Abstract: The emergence of technologies capable of producing real-time data opened new horizons to planning and optimising vehicle routes. Dynamic vehicle routing problems (DVRPs) use real-time information to dynamically calculate the most optimised set of routes. The typical approach is to initially calculate the vehicle routes and dynamically revise them in real-time. This work uses the case study of ASAE, a Portuguese administrative authority specialising in food safety and economic surveillance. The dynamic properties of ASAE’s operational environment are studied, and a solution is proposed to review and efficiently modify the precalculated plan. We propose a weighted utility function based on three aspects: the summed utility of the inspections, the similarity between solutions, and the arrival time. A Disruption Generator generates disruptions on the inspection routes: travel and inspection times, vehicle and inspection breakdowns, utility changes, and unexpected or emerging inspections. We compare the performance of four meta-heuristics: Hill-Climbing (HC), Simulated Annealing (SA), Tabu-Search (TS) and Large neighbourhood Search (LNS). The HC algorithm has the fastest convergence, while SA takes longer to solve the test instances. LNS was the method with higher solution quality, while HC provided solutions with lower utility.
Download

Paper Nr: 250
Title:

Balancing Multiplayer Games across Player Skill Levels using Deep Reinforcement Learning

Authors:

Conor Stephens and Chris Exton

Abstract: The balance or perceived fairness of Level & Character design within multiplayer games depends on the skill level of the players within the game, skills or abilities that have high contributions but require low skill, feel unfair for less skill players and can become the dominant strategy and playstyle if left unchecked. Player skill influences the viable tactics for different map designs, with some strategies only possible for the best players. Level designers hope to create various maps within the game world that are suited to different strategies, giving players interesting choices when deciding what to do next. This paper proposes using deep learning to measure the connection between player skills and balanced level design. This tool can be added to Unity game engine allowing designers to see the impact of their changes on the level’s design on win-rate probability for different skilled teams. The tool is comprised of a neural network which takes as input the level layout as a stacked 2D one hot encoded array alongside the player parameters, skill rating chosen characters; the neural network output is the win rate probability between 0-1 for team 1. Data for this neural network is generated using learning agents that are learning the game using self-play (Silver et al., 2017) and the level data that is used for training the neural network is generated using procedural content generation (PCG) techniques.
Download

Paper Nr: 253
Title:

Molecular Fragments from Incomplete, Real-life NMR Data: Framework for Spectra Analysis with Constraint Solvers

Authors:

Haneen A. Alharbi, Igor Barsukov, Rudi Grosman and Alexei Lisitsa

Abstract: Nuclear Magnetic Resonance (NMR) spectroscopy is a powerful analytical tool that can be used in the elucidation of chemical structures and is widely applied both in academia and industry. Despite using computer-assisted structure elucidation systems, interpretation of NMR data is often laborious, requires high levels of expertise and is not immune to ambiguities. In this multi-disciplinary study, we developed a design of a novel system using a Constraint Satisfaction (CS) framework to utilise unannotated NMR spectra. Additionally, our system allows the utilisation of complementary information obtained/known outside the scope of NMR. Herein we describe a prototype implementation and its empirical evaluation on a set of amino acids, which are a diverse class of important biological compounds. We further employ the CS approach to show the principle limits (ambiguity) of the NMR method in molecular structure elucidation.
Download

Paper Nr: 265
Title:

Systematic Model-based Design of a Reinforcement Learning-based Neural Adaptive Cruise Control System

Authors:

Or A. Yarom, Jannis Fritz, Florian Lange and Xiaobo Liu-Henke

Abstract: In this paper, the systematic model-based design of a reinforcement learning-based neuronal adaptive cruise control is described. Starting with an introduction and a summary of current fundamentals, design methods for intelligent driving functions are presented. The focus is on the first-time presentation of a novel design methodology for artificial neural networks in control engineering. This methodology is then applied and fully validated using the example of an adaptive cruise control system.
Download

Paper Nr: 269
Title:

Momentum Iterative Gradient Sign Method Outperforms PGD Attacks

Authors:

Sreenivasan Mohandas, Naresh Manwani and Durga P. Dhulipudi

Abstract: Adversarial examples are machine learning model inputs that an attacker has purposefully constructed to cause the model to make a mistake. A recent line of work focused on making adversarial training computationally efficient for deep learning models. Projected Gradient Descent (PGD) and Fast Gradient Sign Method (FGSM) are popular current techniques for generating adversarial examples efficiently. There is a tradeoff between these two in terms of robustness or training time. Among the adversarial defense techniques, adversarial training with the PGD is considered one of the most effective ways to achieve moderate adversarial robustness. However, PGD requires too much training time since it takes multiple iterations to generate perturbations. On the other hand, adversarial training with the FGSM takes much less training time since it takes one step to generate perturbations but fails to increase adversarial robustness. Our algorithm achieves better robustness to PGD adversarial training on CIFAR-10/100 datasets and is faster than PGD string adversarial training methods.
Download

Paper Nr: 277
Title:

Old English Universal Dependencies: Categories, Functions and Specific Fields

Authors:

Javier M. Arista

Abstract: The aim of this paper is to lay the foundations of the application of the framework of Universal Dependencies to Old English. Such application will result in the morphological and syntactic annotation of a large data set of Old English with Universal Dependencies categories and relations. The aim of this paper involves two tasks. Firstly, it is necessary to select the relevant categories from the set of universal part-of-speech tags and to identify the Old English exponents of the universal set of morphological features. Secondly, the dependency relations holding in Old English should be listed. The main conclusion of this paper is that two specific fields should be added to the standard Universal Dependencies annotation scheme in order to account to two central aspects of Old English, namely, a gloss field, given the historical character of the language; and a morphological relatedness field, in order to account for its associative lexicon.
Download

Paper Nr: 284
Title:

Evolving Gaussian Mixture Models for Classification

Authors:

Simon Reichhuber and Sven Tomforde

Abstract: The combination of Gaussian Mixture Models and the Expectation Maximisation algorithm is a powerful tool for clustering tasks. Although there are extensions for the classification task, the success of the approaches is limited, in part because of instabilities in the initialisation method, as it requires a large number of statistical tests. To circumvent this, we propose an ’evolutionary Gaussian Mixture Model’ for classification, where a statistical sample of models evolves to a stable solution. Experiments in the domain of Human Activity Recognition are conducted to demonstrate the sensibility of the proposed technique and compare the performance to SVM-based or LSTM-based approaches.
Download

Paper Nr: 289
Title:

Detection of Potential Manipulations in Electricity Market using Machine Learning Approaches

Authors:

Shweta Tiwari, Gavin Bell, Helge Langseth and Heri Ramampiaro

Abstract: Detecting potential manipulations by monitoring trading activities in the electricity market is a time- consuming and challenging task despite the involvement of experienced market surveillance experts. This is due to the increasing complexity of the market structure, contributing to the increase of deceptive anomalous behaviours that can be considered as market abuses. In this paper, we present a novel methodology for detecting potential manipulations in the Nordic day-ahead electricity market by using bid curves data. We first develop a method for processing and reducing the dimensionality of the historical bid curves data using statistical techniques. Then, we train unsupervised machine learning-based models to detect outliers in the pre-processed data. Our methodology captures the sensitivity of the electricity prices resulting from the competitive bidding process and predicts anomalous market behaviours. The results of our experiments show that the proposed approach can complement human experts in market monitoring, by pointing towards relevant cases of manipulation, demonstrating the applicability of the approach.
Download

Paper Nr: 290
Title:

Application of Sequential Neural Networks to Predict the Approximation Degree of Decision-making Systems

Authors:

Jarosław Szkoła and Piotr Artiemjew

Abstract: The paradigm of granular computing appeared from an idea proposed by L. A Zadeh, who assumed that a key element of data mining techniques is the grouping of objects using similarity measures. He assumed that similar objects could have similar decision classes. This assumption also guides other scientific streams such as reasoning by analogy, nearest neighbour method, and rough set methods. This assumption leads to the implication that grouped data, (granules) can be used to reduce the volume of decision systems while preserving their classification efficiency - internal knowledge. This hypothesis has been verified in practice - including in the works of Polkowski and Artiemjew (2007 - 2015) - where they use rough inclusions proposed by Polkowski and Skowron as an approximation tool - using the approximation scheme proposed by Polkowski. In this work, we present the application of sequential neural networks to estimate the degree of approximation of decision systems (the degree of reduction in the size) based on the degree of indiscernibility of the decision system. We use the standard granulation method as a reference method. Pre-estimation of the degree of approximation is an important problem for the considered techniques, in the context of the possibility of their rapid application. This estimation allows the selection of optimal parameters without the need for a classification process.
Download

Paper Nr: 293
Title:

ContourVerifier: A Novel System for the Robustness Evaluation of Deep Contour Classifiers

Authors:

Rania Khalsi, Mallek M. Sallami, Imen Smati and Faouzi Ghorbel

Abstract: DNN certification using abstract interpretation often deals with image-type data, and subsequently evaluates the robustness of the deep classifiers against disturbances on the images such as geometric transformations, occlusion and convolutional noises by modeling them as an abstract domain. In this paper, we propose ContourVerifier, a new system for the evaluation of contour classifiers as we have formulated the abstract domains generated by rigid displacements on contours. This formulation allowed us to estimate the robustness of deep classifiers with different architectures and on different databases. This work will serve as a fundamental building block for the certification of deep models developed for shape recognition.
Download

Area 2 - Agents

Full Papers
Paper Nr: 5
Title:

Towards a Robust, Distributed and Decentralised Smart Energy Management of Microgrids

Authors:

Sandra Garcia-Rodriguez and Hassan A. Sleiman

Abstract: Modern energy systems comprise different entities that interact to allow an intelligent production, distribution, and consumption of energy. They need efficient and distributed demand-response management mechanisms to find optimised configurations of parameters of the grid components. When working with time schedules, optimisation algorithms used for this purpose usually rely on forecasts. However, forecasts bring uncertainty, which is rarely considered in optimisation. This work presents a robust and decentralised optimisation approach that deals also with such uncertainty by searching for optimal power schedule solutions, which are also reliable in unexpected circumstances. Based on message passing, our approach uses meta-heuristics for performing local optimisations. The implementation and validation of our proposal was conducted by means of a distributed multi-agent system, where the obtained results have shown the efficiency of our approach.
Download

Paper Nr: 49
Title:

Coordinated Collision-free Movement of Groups of Agents

Authors:

Jiří Švancara, Marika Ivanová and Roman Barták

Abstract: Coordinating the movement of groups of autonomous agents in a crowded environment is a vital problem with application areas such as warehousing, computer games, or drone art. In this paper, we study the problem of finding collision-free paths for groups of agents such that the groups are kept together like a flock, a fish school, or a military unit. Specifically, we analyze the properties of the problem, propose a SAT formulation based on network flows, and perform a numerical experimental evaluation on various instance types. The results suggest that it is a challenging problem with promising research and application potential. Furthermore, we demonstrate the functionality of our solution method on real educational robots.
Download

Paper Nr: 116
Title:

Reinforcement Learning-based Real-time Fair Online Resource Matching

Authors:

Pankaj Mishra and Ahmed Moustafa

Abstract: Designing a resource matching policy in an open market paradigm is a challenging and complex problem. The complexity is mainly due to the conflicting objectives of the independent resource providers and dynamically arriving online buyers. In specific, providers aim to maximise their revenue, whereas buyers aim to minimise their resource costs. Therefore, to address this complex problem, there is an immense need for a fair matching platform. In specific, the platform must optimise the pricing rule on behalf of resource providers to maximise their revenue at one end. Then, on the other hand, the broker must fairly match the resource request on behalf of buyers. Owing to this we propose a two-step unbiased broker based resource matching mechanism in the auction paradigm. In the first step, the broker computes optimal trade prices on behalf of the providers using a novel reinforcement learning algorithm. Then, in the second step appropriate provider is matched with the buyer’s request based on a novel multi-criteria winner determination strategy. Towards the end, we compare our online resource matching approach with two existing state-of-the-art algorithms. Then, from the experimental results, we show that the novel matching algorithm outperforms the other two baselines.
Download

Paper Nr: 137
Title:

How Does AI Play Football? An Analysis of RL and Real-world Football Strategies

Authors:

Atom Scott, Keisuke Fujii and Masaki Onishi

Abstract: Recent advances in reinforcement learning (RL) have made it possible to develop sophisticated agents that excel in a wide range of applications. Simulations using such agents can provide valuable information in scenarios that are difficult to scientifically experiment in the real world. In this paper, we examine the play- style characteristics of football RL agents and uncover how strategies may develop during training. The learnt strategies are then compared with those of real football players. We explore what can be learnt from the use of simulated environments by using aggregated statistics and social network analysis (SNA). As a result, we found that (1) there are strong correlations between the competitiveness of an agent and various SNA metrics and (2) aspects of the RL agents play style become similar to real world footballers as the agent becomes more competitive. We discuss further advances that may be necessary to improve our understanding necessary to fully utilise RL for the analysis of football.
Download

Paper Nr: 149
Title:

Statistical Model Checking for Probabilistic Temporal Epistemic Logics

Authors:

Yenda Ramesh and M. P. Rao

Abstract: Interpreted Systems and epistemic temporal logics have been employed extensively to study the notion of knowledge in Multi-Agent Systems. New model checking algorithms, as well as adaptations of existing algorithms to this setting have been reported. For the most part, these algorithms have focused on exhaustive state space exploration based approaches. While these approaches yield accurate results to model checking queries, they are often expensive for realistic scenarios. So much so that, many of the applications studied in academic literature deal with small state spaces. In order to scale to real life multi-agent systems with large state spaces, an alternative to exhaustive exploration based techniques is needed. Statistical Model Checking was proposed to alleviate this problem when model checking stochastic systems against temporal logic queries. In this paper, we extend this technique to epistemic temporal logics. The first version of the approach, which we call the vanilla approach, would be to simply generate Monte Carlo samples of the runs of the system and evaluate the query on them. The advantage that SMC is expected to bring is greatly diminished due to the knowledge operator in such systems of logic. For large systems, this would entail an exhaustive exploration of epistemically accessible global states. Our major contribution is to introduce a sampling based approach for the knowledge operator as well. We show that this results in significant performance gains at the expense of a marginal loss in accuracy (1-2% in experimental results) for most epistemic operators. Specifically, we show evidence of a dramatic improvement in time complexity for large Multi-Agent Systems. We substantiate the effectiveness of the approach through case studies that involve a large number of agents.
Download

Paper Nr: 183
Title:

Service Selection for Service-Oriented Architecture using Off-line Reinforcement Learning in Dynamic Environments

Authors:

Yuya Kondo and Ahmed Moustafa

Abstract: Service-Oritented Architeture (SOA) is a style of system design in which the entire system is built from a combination of services, which are functional units of software. The performance of a system designed with SOA depends on the combination of services. In this research, we aim to use reinforcement learning for service selection in SOA. Service selection in SOA is characterized by its dynamic environment and inefficient collection of samples for training. We propose an offline reinforcement learning method in a dynamic environment to solve this problem. In the proposed method, transfer learning is performed by applying fine tuning and focused sampling. Experiments show that the proposed method can adapt to dynamic environments more efficiently than redoing online reinforcement learning every time the environment changes.
Download

Paper Nr: 201
Title:

Multi-objective Risk Analysis for Crowd Evacuation Guidance using Multiple Visual Signs

Authors:

Akira Tsurushima

Abstract: Efficient crowd evacuation guidance is crucial but challenging owing to the randomness involved in evacuation situations and to the unpredictable human behaviors, e.g., herd behavior among evacuees. Many researchers have found that visual evacuation signage is useful for this purpose, and, thus, evacuation guidance systems employing visual signage have been developed. A proper arrangement of visual signs on the premises is necessary to obtain the most out of these attempts; however, several factors make this task challenging, such as multiple conflicting objectives in the evacuations and randomness and uncertainties in the situation. This study formulates the visual evacuation signage assignment problem as a stochastic multi-objective optimization problem and explores the efficient layouts of multiple visual signs on the premises. We consider two objectives for the efficient layout of visual signs, namely, maximizing the number of evacuees selecting the correct exit and minimizing the total evacuation time. The average value at risk is employed to deal with the risks involved in noisy objective functions, while the expected values of these objectives are optimized. Pareto-optimal solutions satisfying both the expected values and the risk measures were explored in cases with one, two and five evacuation signs using the NSGA-II algorithm.
Download

Paper Nr: 213
Title:

Task Handover Negotiation Protocol for Planned Suspension based on Estimated Chances of Negotiations in Multi-agent Patrolling

Authors:

Sota Tsuiki, Keisuke Yoneda and Toshiharu Sugawara

Abstract: We propose a negotiation method that mitigates performance degradation in the multi-agent cooperative patrolling problem not only during planned suspensions for periodic inspection and replacement, but also during the transition period to the suspension. Recent developments in machine and information technologies have led to the expectation of using multiple intelligent agents to control robots. In particular, cooperation between multiple agents is necessary to process tasks that require complex and diverse capabilities or encompass a large environment. Because robots are machines, they need to be regularly inspected and replaced with new ones to prevent unexpected failures and prolong their lifespans. However, suspending agents for such inspections may cause a rapid performance degradation that cannot be neglected in some applications. Such suspensions are usually planned, and the transition period is known in advance, that is, we know which agents will be suspended and when. Our proposed negotiation method allows agents that are scheduled for suspension to hand over important tasks that should not be neglected to other agents. This mitigates the performance degradation during both the transition and suspension periods. The experimental results show that the performance degradation can be significantly reduced compared to existing methods, especially for security surveillance applications.
Download

Paper Nr: 244
Title:

Simulations of a Computational Model for a Virtual Medical Assistant

Authors:

Aryana C. Jackson, Marlène Gilles, Eimear Wall, Elisabetta Bevacqua, Pierre De Loor and Ronan Querrec

Abstract: We propose a virtual medical assistant to guide both novice and expert caregivers through a procedure without the direct help of medical professionals. Our medical assistant uses situational leadership to handle all interaction with a caregiver, which works by identifying the readiness level of the caregiver in order to match them with an appropriate style of communication. The agent system (1) obtains caregiver behavior during the procedure, (2) calculates a readiness level of the caregiver using that behavior, and (3) generates appropriate agent behavior to progress the procedure and maintain a positive interaction with the caregiver.
Download

Paper Nr: 252
Title:

Speech Perception and Implementation in a Virtual Medical Assistant

Authors:

Aryana C. Jackson, Yann Glémarec, Elisabetta Bevacqua, Pierre De Loor and Ronan Querrec

Abstract: In emergency medical procedures, positive and trusting interactions between followers and leaders are imperative. That interaction is even more important when a virtual agent assumes the leader role and a human assumes the follower role. In order to manage the human-computer interaction, situational leadership is employed to match the human follower to an appropriate leadership style embodied by the agent. Situational leadership was used to create 33 utterances indicative of the four different leadership styles. A participant evaluation was then carried out in order to examine (1) whether perceptions of leader trust and motivation vary dependent on both readiness level and utterance syntax and (2) whether follower ability and willingness are affected by the leader’s speech. We found that general perceptions of leadership behavior influenced follower performance and that the leader’s speech influences followers’ ability. Finally, we demonstrate how the results of this study are implemented in a virtual agent system.
Download

Short Papers
Paper Nr: 9
Title:

Towards Multi-agent Reinforcement Learning using Quantum Boltzmann Machines

Authors:

Tobias Müller, Christoph Roch, Kyrill Schmid and Philipp Altmann

Abstract: Reinforcement learning has driven impressive advances in machine learning. Simultaneously, quantum-enhanced machine learning algorithms using quantum annealing underlie heavy developments. Recently, a multi-agent reinforcement learning (MARL) architecture combining both paradigms has been proposed. This novel algorithm, which utilizes Quantum Boltzmann Machines (QBMs) for Q-value approximation has outperformed regular deep reinforcement learning in terms of time-steps needed to converge. However, this algorithm was restricted to single-agent and small 2x2 multi-agent grid domains. In this work, we propose an extension to the original concept in order to solve more challenging problems. Similar to classic DQNs, we add an experience replay buffer and use different networks for approximating the target and policy values. The experimental results show that learning becomes more stable and enables agents to find optimal policies in grid-domains with higher complexity. Additionally, we assess how parameter sharing influences the agents’ behavior in multi-agent domains. Quantum sampling proves to be a promising method for reinforcement learning tasks, but is currently limited by the Quantum Processing Unit (QPU) size and therefore by the size of the input and Boltzmann machine.
Download

Paper Nr: 22
Title:

Allocation Considering Agent Importance in Constrained Robust Multi-Team Formation

Authors:

Ryo Terazawa and Katsuhide Fujita

Abstract: Forming a team to accomplish a given mission is a significant challenge in multi-agent systems. The mission-oriented team formation problem is selecting agents with a set of skills to accomplish a mission. Each mission can be accomplished by assigning a team with the necessary skills. Furthermore, the team needs to accomplish the mission, even if an agent fails in an environment where agents are lost between missions. In addition, other aspects besides the ability to accomplish the mission are considered in team formation. In this paper, we focus on the mission-oriented constrained robust multi-team formation problem. We define a framework, decision, and several optimization problems. Then, we propose an algorithm for the optimization problem with fixed robustness. In our experiments, we confirmed the search efficiency by changing the order in which agents are searched. The results show that a short runtime is obtained in searching for agents with high importance when the ratio of solutions to the search space is large. Furthermore, the runtime is minimized in searching for easily pruned agents when the ratio of solutions to the search space is small.
Download

Paper Nr: 51
Title:

Online Inference of Robot Navigation Parameters from a Semantic Map

Authors:

Benjamin Kisliuk, Christoph Tieben, Nils Niemann, Christopher Bröcker, Kai Lingemann and Joachim Hertzberg

Abstract: Agriculture is becoming one of the key application fields for mobile robots. At the same time it poses serious challenges for true autonomous systems due to its heterogeneous and dynamic nature. To act robustly and reliably, robotic behaviour needs to be controlled by an intelligence, making explainable and informed decisions based on knowledge of its surroundings. However, this knowledge cannot only be derived from sensor data but has to be based on prior knowledge and external sources as well to comprehensively represent a robots deployment site. By representing this knowledge in formal and thus machine readable way, automated inference improves the handling of the complex nature of these requirements. In this paper, we show how quantitative and qualitative control parameters regarding a mobile robots navigation can be derived from a manually modelled semantic map of an agricultural deployment site. Also we describe how such a system can be integrated into a typical ROS system architecture. By making the derived knowledge easily available, the robotic system is enabled to dynamically adapt route planning on an agricultural deployment site and to switch between different local planning algorithms according to situational and prior knowledge.
Download

Paper Nr: 54
Title:

NC4OMAS: A Norms-based Approach for Open Multi-Agent Systems Controllability

Authors:

Mohamed Sedik Chebout, Farid Mokhati and Mourad Badri

Abstract: Normative Open Multi-Agent Systems (NOMAS) are systems in which norms play a crucial role for organizing, coordinating, controlling agents’ behaviours and interactions. In addition, heterogeneous agents, in Open Multi-Agent Systems (OMAS), can work in similar or different ends. This might deviate the target system state and lead to a chaotic behaviour. A particular kind of OMAS is implemented based on AGR (Agent/Group/Role) model. This paper proposes a novel Norms-based Controllability approach for AGR-based OMAS (NC4OMAS). Mainly, the proposed approach is divided into two phases: monitoring and controlling. Aspect-Oriented Programming (AOP) technique is used for norm monitoring compliance. Also, JAVA Expert System Shell (JESS) is used for norm specification, norm modification and for making inference over norms at runtime. In order to address limitations and advantages of our approach, we summarise the most relevant works on norms-based control according to some comparison criteria we proposed.
Download

Paper Nr: 73
Title:

Machine-learned Behaviour Models for a Distributed Behaviour Repository

Authors:

Alexander Jahl, Harun Baraki, Stefan Jakob, Malte Fax and Kurt Geihs

Abstract: Dynamically organised multi-agent systems that consist of heterogeneous participants require cooperation to fulfil complex tasks. Such tasks are commonly subdivided into subtasks that have to be executed by individual agents. The necessary teamwork demands coordination of the involved team members. In contrast to typical approaches like agent-centric and organisation-centric views, our solution is based on the task-centric view and thus contains active task components which select agents focusing on their Skills. It enables an encapsulated description of the task flow and its requirements including team cooperation, organisation, and location-independent allocation processes. Besides agent properties that represent syntactical and semantic information, agent behaviours are considered as well. The main contributions of this paper are hyperplane-based machine-learned Behaviour Models that are generated to capture the behaviour and consider the Behaviour Implementations as black boxes. These Behaviour Models are provided by a distributed behaviour repository that enables tasks to actively select fitting Behaviour Implementations. We evaluated our approach based on agents playing chessboard-like games autonomously.
Download

Paper Nr: 90
Title:

“Robot Steganography”: Opportunities and Challenges

Authors:

Martin Cooney, Eric Järpe and Alexey Vinel

Abstract: Robots are being designed to communicate with people in various public and domestic venues in a perceptive, helpful, and discreet way. Here, we use a speculative prototyping approach to shine light on a new concept of robot steganography (RS): that a robot could seek to help vulnerable populations by discreetly warning of potential threats: We first identify some potentially useful scenarios for RS related to safety and security– concerns that are estimated to cost the world trillions of dollars each year–with a focus on two kinds of robots, a socially assistive robot (SAR) and an autonomous vehicle (AV). Next, we propose that existing, powerful, computer-based steganography (CS) approaches can be adopted with little effort in new contexts (SARs), while also pointing out potential benefits of human-like steganography (HS): Although less efficient and robust than CS, HS represents a currently-unused form of RS that could also be used to avoid requiring a computer to receive messages, detection by more technically advanced adversaries, or a lack of alternative connectivity (e.g., if a wireless channel is being jammed). Some unique challenges of RS are also introduced, that arise from message generation, indirect perception, and effects of perspective. Finally, we confirm the feasibility of the basic concept for RS, that messages can be hidden in a robot’s behaviors, via a simplified, initial user study, also making available some code and a video. The immediate implication is that RS could potentially help to improve people’s lives and mitigate some costly problems, as robots become increasingly prevalent in our society–suggesting the usefulness of further discussion, ideation, and consideration by designers.
Download

Paper Nr: 98
Title:

LIMP: Incremental Multi-agent Path Planning with LPA

Authors:

Mücahit Alkan Yorgancı, Fatih Semiz and Faruk Polat

Abstract: The multi-agent pathfinding (MAPF) problem is defined as finding conflict-free paths for more than one agent. There exist optimal and suboptimal solvers for MAPF, and most of the solvers focus on the MAPF problem in static environments, but the real world is far away from being static. Motivated by this requirement, in this paper, we introduce an incremental algorithm to solve MAPF. We focused on discrete-time and discrete space environments with the unit cost for all edges. We proposed an algorithm called incremental multi-agent path planning with LPA* (LIMP) and discrete lifelong planning A* (DLPA*) for solving I-MAPF (Incremental MAPF). LIMP is the combination of two algorithms which are the Conflict Based Search D*-lite (CBS-D*- lite) (Semiz and Polat, 2021) and DLPA*. DLPA* is just a tailored version of the lifelong planning A* (Koenig et al., 2004) which is an incremental search algorithm for one agent. We have shown that LIMP outperforms Conflict Based Search replanner (CBS-replanner) and CBS-D*-lite (Semiz and Polat, 2021) in terms of speed. Moreover, in terms of cost, LIMP and CBS-D*-lite perform similarly, and they are close to CBS-replanner.
Download

Paper Nr: 104
Title:

Operationalizing Behavior Change Techniques in Conversational Agents

Authors:

Maria I. Bastos, Ana P. Cláudio, Isa B. Félix, Mara P. Guerreiro, Maria B. Carmo and João Balsa

Abstract: Departing from previous work on the use of well-established behavior change techniques in an mHealth intervention based on a conversational agent (CA), we propose in this contribution a new architecture for the design of behavior change CAs. This novel approach combines the use of an advanced natural language platform (Dialogflow) with the explicit representation, in an ontology, of how behavior change techniques can be operationalized. The integration of these two components is explained, as well as the most challenging aspect of using the advanced features of the platform in a way that allowed the agent to lead the dialogue flow, when needed. A successful proof of concept was built, which can be the basis for the development of advanced conversational agents, combining natural language tools with ontology-based knowledge representation.
Download

Paper Nr: 117
Title:

Narrative Economics of the Racetrack: An Agent-Based Model of Opinion Dynamics in In-play Betting on a Sports Betting Exchange

Authors:

Rasa Guzelyte and Dave Cliff

Abstract: We present first results from a new agent-based model (ABM) of a sports-betting exchange (such as those operated by BetFair, BetDdaq, and SMarkets, among other companies) in which each agent holds a dynamically- varying opinion about some uncertain future event (such as which competitor will win a particular horse race) and in which all agents interact with the betting exchange to find counterparties holding an opposing view with whom they can then enter into a bet with. We extend methods from Opinion Dynamics (OD) research to give each agent an opinion at any particular time which is influenced partially by local interactions with other agents (as is common in the OD literature), partially by globally available information (as published to all by the betting exchange) and partially by the progressive reduction in uncertainty in the system (i.e., eventually all agents know which horse has won the race). Our work here is motivated by the prize-winning ICAART2021 paper of Lomas & Cliff, who integrated OD methods with ABMs of financial markets to explore issues in Narrative Economics, an approach recently proposed and popularised by Nobel Laureate Robert Shiller, but here we explore a significantly different type of market: a betting market (which has strong similarities to a financial market for tradeable derivative contracts such as futures or options). The novel contributions of this paper are centred on the extension of OD methods to situations in which there is a mix of local and global influence, and in which uncertainty progressively reduces to zero. We present results from our initial proof-of-concept implementation. The Python source-code for our ABM is freely available on Github for other researchers to replicate and extend the work reported here.
Download

Paper Nr: 122
Title:

Coordination Mechanisms with Misinformation

Authors:

Constantinos Varsos, Michail Fasoulakis, Giorgos Flouris and Marina Bitsaki

Abstract: We introduce a novel approach for coordination mechanisms in games, based on the idea of misinforming players about the game formulation in order to steer them towards a behaviour that leads to an improved outcome in terms of social welfare. As a use case, we study single-commodity non-atomic congestion games with parallel links and affine cost functions. We propose a simple mechanism that provides to the players the right incentives to adopt a socially optimal behaviour by misinforming them with regards to the latency functions of the links, under various assumptions. We use a metric called the Price of Misinformation to quantify the effect of misinformation on social welfare (compared to the optimum of the actual game), and show that our mechanism can minimise this metric, resulting in values that are better than the Price of Anarchy (i.e., the social outcome without any intervention from the designer).
Download

Paper Nr: 127
Title:

Machine-checked Verification of Cognitive Agents

Authors:

Alexander B. Jensen

Abstract: The ability to demonstrate reliability is an important aspect in deployment of software systems. This applies to cognitive multi-agent systems in particular due to their inherent complexity. We are still pursuing better approaches to demonstrate their reliability. The use of proof assistants and theorem proving has proven itself successful in verifying traditional software programs. This paper explores how to apply theorem proving to verify agent programs. We present our most recent work on formalizing a verification framework for cognitive agents using the proof assistant Isabelle/HOL.
Download

Paper Nr: 138
Title:

A Mechanism for Multi-unit Multi-item Commodity Allocation in Economic Networks

Authors:

Pankaj Mishra, Ahmed Moustafa and Fenghui Ren

Abstract: n this work, we introduce a novel resource allocation mechanism that aims to maximise the social welfare of the market in procurement auctions. Specifically, we consider a market setting with multiple units of homogeneous resources. In such settings, buyers submit their resource requests to a limited number of known providers. This limited number of providers might in turn lead to a provider monopoly in the market and a scarcity of the resources. To address this problem, we propose a novel information diffusion-based resource allocation mechanism for resource allocation in procurement auctions. The proposed mechanism focuses on procuring multiple units of homogeneous resources. In this regard, the proposed mechanism incentivises the providers to truthfully diffuse the procurement information to their neighbours. This information diffusion aids the buyers to procure the required amounts of commodities/resources at the minimum possible prices. In addition, the proposed mechanism gives fair chances to the distant providers to fairly participate in the procurement auction. Further, we prove that the proposed mechanism minimises the procurement costs, with no deficits, compared to the Vickrey-Clarke-Groves mechanism. Finally, based on the experiments, we show that the proposed mechanism has comparatively lesser procurement costs.
Download

Paper Nr: 142
Title:

Highways in Warehouse Multi-Agent Path Finding: A Case Study

Authors:

Vojtěch Rybář and Pavel Surynek

Abstract: Orchestrating warehouse sorting robots each transporting a single package from the conveyor belt to its destination is a NP-hard problem, often modeled Multi-agent path-finding (MAPF) where the environment is represented as a graph and robots as agents in vertices of the graph. However, in order to maintain the speed of operations in such a setup, sorting robots must be given a route to follow almost at the moment they obtain the package, so there is no time to perform difficult offline planning. Hence in this work, we are inspired by the approach of enriching conflict-based search (CBS) optimal MAPF algorithm by so-called highways that increase the speed of planning towards on-line operations. We investigate whether adding highways to the underlying graph will be enough to enforce global behaviour of a large number of robots that are controlled locally. If we succeed, the slow global planning step could be omitted without significant loss of performance.
Download

Paper Nr: 146
Title:

Supporting the Adaptation of Agents' Behavioral Models in Changing Situations by Presentation of Continuity of the Agent's Behavior Model

Authors:

Yoshimasa Ohmoto, Junya Karasaki and Toyoaki Nishida

Abstract: In this study, we attempted to make participants continually estimate an agent’s behavioral model by having the agent itself present the continuity of its behavior model during a task. By doing so, we aimed to encourage users to pay attention to changes in the agent’s behavioral model and to make the user continuously change the relationship between themselves and the agent. In order to make the participants continually estimate the agent’s behavioral model, we proposed the method of “presentation of continuity of the agent’s behavior model.” We implemented agents based on this and conducted an experiment using an animal-guiding task in which one human and two agents cooperated. As a result, we were able to significantly increase the degree to which participants paid attention to the agents and induce active interaction behavior. This suggests that the proposed method contributed to maintaining a relationship between the agents and the participant even in changing situations.
Download

Paper Nr: 193
Title:

On Combining Domain Modeling and Organizational Modeling for Developing Adaptive Cyber-Physical Systems

Authors:

Jan Sudeikat and Michael Köhler-Bußmeier

Abstract: Cyber-physical Systems (CPS) integrate physical and computational entities into coherent Systems of Systems. Since CPS are typically both socio-technical and embedded in a physical environment, these systems require adaptive properties, e.g. in order to respond to environmental changes. The integration of Multi-Agent Systems (MAS) in industrial and CPS is an active research topic. In this paper, we outline our work in progress on utilizing, combining and supplementing established modeling approaches, i.e. domain specific (reference) architecture models and organizational MAS models for the development of decentralized, adaptive CPS.
Download

Paper Nr: 194
Title:

Impact of Error-making Peer Agent Behaviours in a Multi-agent Shared Learning Interaction for Self-Regulated Learning

Authors:

Sooraj Krishna and Catherine Pelachaud

Abstract: Agents in a learning environment can have various roles and social behaviours that can influence the goals and motivation of the learners in distinct ways. Self-regulated learning (SRL) is a comprehensive conceptual framework that encapsulates the cognitive, metacognitive, behavioural, motivational and affective aspects of learning and entails the processes of goal setting, monitoring progress, analyzing feedback, adjustment of goals and actions by the learner. The study aims to understand how error-making behaviours in the peer agent role would influence the learner perceptions of agent roles, related behaviours and self-regulation. We present a multi-agent learning interaction involving the pedagogical agent roles of tutor and peer learner defined by their social attitudes and competence characteristics, delivering specific regulation scaffolding strategies for the learner. The results from the study suggests the effectiveness of error-making behaviours in peer agent for clearly establishing the pedagogical roles in a multi-agent learning interaction context along with significant influences in self-regulation and agent competency perceptions in the learner.
Download

Paper Nr: 195
Title:

SMT-based BMC for Dense Timed Interpreted Systems and EMTLK Properties

Authors:

Agnieszka M. Zbrzezny, Andrzej Zbrzezny and Bożena Woźna-Szcześniak

Abstract: The use of automated verification, performed by the analysis of their models, is often recommended to assess the correctness of safety-critical systems, failure of which could cause dramatic consequences for both people and hardware. In the past, several automated verification methods, including model checking, have been proposed and consequently applied for the trustworthy development of real-time multi-agent systems (RTMAS). In this paper, we investigate a Satisfiability Modulo Theories based Bounded Model Checking (SMT-BMC) method for EMTLK (the existential fragment of an epistemic Metric Temporal Logic) that is interpreted over models generated by Dense Timed Interpreted Systems (DTIS). In particular, we translate the existential model checking problem for EMTLK to the existential model checking problem for a variant of an epistemic Linear Temporal Logic with a new set of propositional variables (called ELTLKq), and we provide an SMT-BMC technique for ELTLKq. We have implemented our technique and tested it using the Timed Generic Pipeline Paradigm scenario. Our preliminary experimental results allow us to draw positive conclusions regarding the future applications of our new method in the automated verification of other benchmarks for RTMAS modelled by DTIS.
Download

Paper Nr: 203
Title:

An Agent-based Model Study on Subsidy Fraud in Technological Transition

Authors:

Hao Yang, Xifeng Wu, Sijia Zhao, Hatef Madani, Jin Chen and Yu Chen

Abstract: The evolution of a society is inextricably linked to technological transition, which is based on both innovation and dissemination of technologies. To protect the vulnerable new generation of technology, government subsidies are one of the most common and effective tools. However, not all subsidy policies can lead to a healthy development of market shares. Subsidy fraud is one of the most problematic issues that can arise under an imperfect system. This paper identifies an interesting subsidy fraud like phenomenon via a validated agent-based model. After analysing the mechanism of the transition of technology in the model, we drive the condition upon which subsidy fraud could occur.
Download

Paper Nr: 217
Title:

Framework to Predict Energy Prices and Trades in the Wholesale Market of PowerTAC

Authors:

Filipe P. Reis, Helena V. Ferreira and Ana P. Rocha

Abstract: Machine and Deep Learning techniques have been widely used in the PowerTAC competition to forecast the price of energy as a bulk, amongst other ends. In order to allow agents to quickly set up, train, and test python-built models, we developed a framework based on a micro-service architecture suitable for predicting wholesale market prices in PowerTAC. The architecture allows for algorithms to be implemented in Python as opposed to the language used in PowerTAC, Java. This paper also presents two datasets, one for the task of classifying whether trades occur, and another for the task of predicting the clearing price of trades that occur. We benchmark these results with basic methods like linear regression, random forest, and a neural network.
Download

Paper Nr: 220
Title:

Quantifying Multimodality in World Models

Authors:

Andreas Sedlmeier, Michael Kölle, Robert Müller, Leo Baudrexel and Claudia Linnhoff-Popien

Abstract: Model-based Deep Reinforcement Learning (RL) assumes the availability of a model of an environment’s underlying transition dynamics. This model can be used to predict future effects of an agent’s possible actions. When no such model is available, it is possible to learn an approximation of the real environment, e.g. by using generative neural networks, sometimes also called World Models. As most real-world environments are stochastic in nature and the transition dynamics are oftentimes multimodal, it is important to use a modelling technique that is able to reflect this multimodal uncertainty. In order to safely deploy such learning systems in the real world, especially in an industrial context, it is paramount to consider these uncertainties. In this work, we analyze existing and propose new metrics for the detection and quantification of multimodal uncertainty in RL based World Models. The correct modelling & detection of uncertain future states lays the foundation for handling critical situations in a safe way, which is a prerequisite for deploying RL systems in real-world settings.
Download

Paper Nr: 229
Title:

Toward Crowdsourced Knowledge Graph Construction: Interleaving Collection and Verification of Triples

Authors:

Helun Bu and Kazuhiro Kuwabara

Abstract: This paper presents a method for building a knowledge graph using crowdsourcing. The collection and verification of pieces of knowledge are essential components of building a high-quality knowledge graph. We introduce fill-in-the-blank-type of quizzes to collect knowledge as triples and true-or-false-type quizzes to verify the collected triples. We also present score functions to evaluate and select a quiz for efficient knowledge graph construction based on the workers’ past inputs. The collection and verification processes are dynamically interleaved using weights in the score function. Simulation results show that the proposed approach can collect and verify distributed knowledge among casual workers.
Download

Paper Nr: 240
Title:

Completion of User Preference based on CP-nets in Automated Negotiation

Authors:

Jianlong Cai, Jieyu Zhan and Yuncheng Jiang

Abstract: Automated negotiation is a process in which autonomous agents negotiate with the opponents to achieve some specific purposes for their users, such as maximising the users’ benefits. CP-net is one of the most important representations of user preferences in automated negotiation due to its ability and flexibility to express interdependent relationship among issues. In order to be able to negotiate better on behalf of users, a negotiating agent needs to fully understand its user’s preferences, so that it can adopt suitable negotiation strategies and obtain ideal negotiation results. However, the preference information provided to the negotiating agents by the users is often incomplete. Hence, based on partial preference information provided, this paper proposes a module in negotiation framework to complete user total preferences that are represented by CP-nets. The experimental results show the validity of CP-nets structure learning algorithm in the proposed module and confirm that the module can help users achieve better agreements in negotiation.
Download

Paper Nr: 257
Title:

Design of an Autonomous Distributed Multi-agent Mission Control System for a Swarm of Satellites

Authors:

Petr Skobelev, Gennady Myatov, Vladimir Galuzin, Anastasia Galitskaya, Anton Ivanov and Aleksandr Chernyavskii

Abstract: The paper describes an autonomous distributed multi-agent system for mission control of a multi-satellite swarm, using direct data exchange between satellites in space via a radio channel to make coordinated collective decisions. The main advantage of autonomous control on board the vehicle is the ability to use the current data on its state to quickly respond to events in real time, without having to wait for a response or instructions from the Earth. The proposed approach develops the principles of creating self-organizing systems and is supposed to be implemented in several stages of the space mission. The first stage consists in conducting experiments on the use of inter-satellite interaction in order to assess and clarify the possibility of performing and correcting the plan of operations built on the ground with account of the current telemetry data obtained in real time. At the second stage, it is planned to use more powerful on-board computers and organize fully autonomous control in a mesh network formed by the satellites for a distributed solution of the observation problem, surveying a given area in the interests of ecology and solving other problems requiring coordinated interaction of devices. In this regard, this paper presents a refined brief problem statement for planning the work of a multi-satellite swarm in relation to the previously considered one. A brief description of the developed system is given, which makes it possible to implement processing applications for performing space experiments by means of the ground circuit and resources of the space constellation. The paper also presents the structure and functions of the autonomous multi-agent system and protocols of agent interaction, as well as models and methods of multi-agent group management. Prospects for further development and practical application of the approach are discussed.
Download

Paper Nr: 262
Title:

Study on Applying Decentralized Evolutionary Algorithm to Asymmetric Multi-objective DCOPs with Fairness and Worst Case

Authors:

Toshihiro Matsui

Abstract: The Distributed Constraint Optimization Problem (DCOP) is a fundamental research area on cooperative problem solving in multiagent systems. An extended class of DCOPs represents a situation where each agent locally evaluates its partial problem with its individual constraints and objective functions on the variables shared by neighboring agents. This is a multi-objective problem on the preference of individual agents, and a set of aggregation and comparison operators is employed for a metric of social welfare among the agents. We concentrate on the case of social welfare criteria based on leximin/leximax that captures fairness among agents. Since the constraints in the practical settings of asymmetric multi-objective DCOPs are too dense for exact solution methods, scalable but inexact solution methods are necessary. We focus on employing a version of an evolutionary algorithm called AED which was designed for the original class of DCOPs. We apply the AED algorithm to asymmetric multi-objective DCOPs to handle asymmetry. We also replace the criteria in the sampling process by one of the social welfare criteria and experimentally investigate the sampling criteria in the search process.
Download

Paper Nr: 275
Title:

Negotiation in Ride-hailing between Cooperating BDI Agents

Authors:

Ömer I. Erduran, Marcel Mauri and Mirjam Minor

Abstract: These days, ride-hailing is an emerging trend in Mobility as a service (MaaS). First services involving human taxi drivers such as Uber, Lyft and DiDi are commercially successful. With the rise of autonomous vehicles, self-organized fleets for ride-hailing systems come into the focus of research. Multi-agent systems (MAS) provide solutions for many challenges of this application scenario. Especially, the communication of cooperating agents is beneficial for a structured and well planned task distribution. In this paper, we investigate a MAS for autonomous vehicles in MaaS and put the focus on a negotiation based assignment of customer trips. An agent model concept is introduced where the main type, the vehicle agent is designed following a BDI architecture. The communication system for the MAS is implemented by using the contract net protocol. We develop the negotiation process and furthermore evaluate the agent communication with respect to its impact on pickup time satisfaction and environmental sustainability using two quality measures, which calculate the average travel distance and the order dropout rate. An experimental setup including historical trip data in a simulation demonstrates the feasibility of our approach.
Download

Paper Nr: 25
Title:

Multi Robot Surveillance and Planning in Limited Communication Environments

Authors:

Vibhav I. Kedege, Aleksander Czechowski, Ludo Stellingwerff and Frans. A. Oliehoek

Abstract: Distributed robots that survey and assist with search & rescue operations usually deal with unknown environments with limited communication. This paper focuses on distributed & cooperative multi-robot area coverage strategies of unknown environments, having constrained communication. Due to restricted communication there is performance loss for the multi-robot team, in terms of increased number of steps to cover an area. From simulation results, it is shown that enabling partial communication amongst robots can recover a significant amount of performance by decreasing the number of steps required for area coverage. Additionally it is found that partially communicating robots that predict the paths of peers do not perform significantly different from robots that are only partially communicating. This is found due to predictions spreading the robots away from one another, which reduces meeting times and instances of inter-robot data sharing.
Download

Paper Nr: 27
Title:

An Interactive Environment to Support Agent-based Graph Programming

Authors:

Daniel Blashaw and Munehiro Fukuda

Abstract: We apply agent-based modeling (ABM) to distributed graph analysis where a large number of reactive agents roam over a distributed graph to find its structural attributes, (e.g., significant subgraphs including triangles in a social network and network motifs in a biological network). Of importance is providing data scientists with an interactive environment to support agent-based graph programming, which enables interactive verification of agent behaviors, trial-and-error operations, and visualization of graphs and agent activities. This paper presents and evaluates our implementation techniques of these interactive features.
Download

Paper Nr: 56
Title:

LSTM-based Abstraction of Hetero Observation and Transition in Non-Communicative Multi-Agent Reinforcement Learning

Authors:

Fumito Uwano

Abstract: This study focuses on noncommunicative multiagent learning with hetero-information where agents observe each other in different resolutions of information. A new method is proposed for adapting the time dimension of the hetero-information from the observation by expanding the Asynchronous Advantage Actor–Critic (A3C) algorithm. The profit minimizing reinforcement learning with oblivion of memory mechanism was the previously used noncommunicative and cooperative learning method in multiagent reinforcement learning. We then insert an long short-term memory (LSTM) module into the A3C neural network to adapt to the time dimension influence of the hetero-information. The experiments investigate the performance of the proposed method on the hetero-information environment in terms of the effectiveness of LSTM. The experimental results show that: (1) the proposed method performs better than A3C. Without the LSTM module, the proposed method enabled the agents’ learning to converge. (2) LSTM can adapt the time dimension of the input information.
Download

Paper Nr: 66
Title:

Requirements Engineering Challenges and Techniques in Building Chatbots

Authors:

Geovana S. Silva and Edna D. Canedo

Abstract: The proper execution of requirements engineering processes can be decisive for the success of software development and, for chatbots, it is no different. Chatbots have been gaining a lot of space, especially in customer service. Requirements engineering processes for chatbots are as hard to perform as for any other machine-learning system and the conversational nature makes it even harder. Taking this into consideration, in this work we survey chatbot practitioners to unveil the requirements elicitation and documentation techniques they have been using in the industry, besides the challenges they face while going through this process. Responses show that elicitation techniques are not much far from techniques used in other fields, but for documentation techniques that are new forms of documentation such as conversation flows. Moreover, meeting stakeholder’s requirements and managing information exchange are their biggest challenges in eliciting and documenting chatbot requirements.
Download

Paper Nr: 134
Title:

An Inspection Technique Proposal for the Verification of Requirements Specification Documents for Multi-Agent Systems

Authors:

Giovane D. Mendonça, Gilleanes A. Guedes and Iderli S. Filho

Abstract: Requirements engineering is an important area of software engineering dedicated to eliciting, analysing, specifying, and validating software requirements to ensure the correct understanding of what needs to be developed. The requirements specification objective is to provide a detailed description of what the software must do, it involves the production of a document that can be systematically reviewed, evaluated, and approved. Problems in the requirements are appointed among the main causes of failures in software projects. Therefore, performing requirements verification and validation is extremely important to ensure the software quality. Multi-Agent systems are a type of software with particular requirements, beyond the commonly found among other systems, since they are composed by several autonomous and proactive agents that divide the problem to be solved among them. Thus, requirements engineering needs to be adapted for this kind of system and the produced documents need to be verified too. Several techniques were proposed for requirements inspection, among them there is the Perspective-Based Reading. However, as in the other approaches, this technique does not allow the inspection of particular requirements for multi-agent systems. Taking this in consideration, our work has as its objective to adapt this technique so as to allow the verification of requirements specification documents specific for this kind of system.
Download

Paper Nr: 143
Title:

Multi-agent Policy Gradient Algorithms for Cyber-physical Systems with Lossy Communication

Authors:

Adrian Redder, Arunselvan Ramaswamy and Holger Karl

Abstract: Distributed online learning over delaying communication networks is a fundamental problem in multi-agent learning, since the convergence behaviour of interacting agents is distorted by their delayed communication. It is a priori unclear, how much communication delay can be allowed, such that the joint policies of multiple agents can still converge to a solution of a multi-agent learning problem. In this work, we present the decentralization of the well known deep deterministic policy gradient algorithm using a communication network. We illustrate the convergence of the algorithm and the effect of lossy communication on the rate of convergence for a two-agent flow control problem, where the agents exchange their local information over a delaying wireless network. Finally, we discuss theoretical implications for this algorithm using recent advances in the theory of age of information and deep reinforcement learning.
Download

Paper Nr: 150
Title:

Model Analysis of Human Group Behavior Strategy using Cooperative Agents

Authors:

Norifumi Watanabe and Kota Itoda

Abstract: Flexible and cooperative human group behavior are realized by changing our intentions and behaviors based on dynamic estimation of other participants’ intention, and also adjustment of self and others’ intention. We analyze human small group behavior using cooperative pattern task in 2D grid world to clarify an individual action selection process including inference of others’ intention and adjustment of intention among participants. In previous research, we have constructed behavior strategy models based on the human behavioral experiments, implemented the models to cooperative agents, and confirmed the goal achievement in almost the same steps to humans in the agent simulations. In this research, we analyze combinations of human behavior strategies realizing group behavior by comparing agent behavior to subjects behavior.
Download

Paper Nr: 176
Title:

Tackling Train Routing via Multi-agent Pathfinding and Constraint-based Scheduling

Authors:

Jiří Švancara and Roman Barták

Abstract: The train routing problem deals with allocating railway tracks to trains so that the trains follow their timetables and there are no collisions among the trains (all safety rules are followed). This paper studies the train routing problem from the multi-agent pathfinding (MAPF) perspective, which proved very efficient for collision-free path planning of multiple agents in a shared environment. Specifically, we modify a reduction-based MAPF model to cover the peculiarities of the train routing problem (various train lengths, in particular), and we also propose a new constraint-based scheduling model with optional activities. We compare the two models both theoretically and empirically.
Download

Paper Nr: 182
Title:

Risk-oriented Behavior Design for Traffic Simulation

Authors:

Philippe Mathieu and Antoine Nongaillard

Abstract: With the advent of the autonomous vehicle and the transformation of the automobile sector in the next decade, road traffic simulation has taken off again, and behavioral testing represents a significant area. A collision is a potentially complex phenomenon that is very difficult to study. In most tools, collisions are predefined phenomena, preventing the study of behavioral factors’ impacts on these collisions. The notion of risk-taking is essential in individual driving behavior to obtain realistic traffic at both the macroscopic (flow) and microscopic (individual behavior) levels. We propose a model where collisions are unpredictable emerging phenomena resulting from individual deterministic behaviors where risk-taking parameters ease the design of various behaviors.
Download

Paper Nr: 188
Title:

Agent-based Modeling for Dynamic Hitchhiking Simulation and Optimization

Authors:

Corwin Fèvre, Hayfa Zgaya-Biau, Philippe Mathieu and Slim Hammadi

Abstract: Although many new transportation services have emerged, hitchhiking continues to be popular, especially in rural areas. In the last 10 years, many countries have tried to encourage and revitalize this mode of transport for its ecological and social aspects. The objective is then to develop tools to ensure the connection of the users as well as the optimization of their journey while respecting the dynamic and volatile character of hitchhiking. In this perspective, we propose the Realtime Trip Avaibility Graph (ReTAG) approach. This approach consists of a recursive algorithm to identify and filter the relevant drivers for the riders. This algorithm generates a graph that allows the riders to establish a perception of the set of rideshares that are eligible and profitable to their situation. We establish a multi-agent system to describe the behavior and interactions of hitchhikers and drivers. We propose a comparative study of two hitchhiker behaviors. The first one simulating the behavior of a real hitchhiker, i.e. without any knowledge of his environment. The second one simulating a hitchhiker connected to an information system, and thus with knowledge of a part of the environment. We compare these two behaviors on more or less challenging problem instances in order to have a panel of convincing results. We conclude that the connected hitchhiker is superior to the real hitchhiker on a set of indicators such as the waiting time and the instance resolution speed.
Download

Paper Nr: 241
Title:

Incorporating Communicative Patterns into Ebdi Agents

Authors:

Yanet Sánchez, Teresa Coma, Antonio Aguelo and Eva Cerezo

Abstract: This work proposes the use of the well-known Satir communicative patterns in Emotional Belief-Desires- Intentions frameworks (EBDIs) aimed to support the management of Embodied Conversational Agents (ECAs). It shows how to include Satir’s model into the ABC-EBDI framework. The framework is based on the ABC psychological model and considers, not only the behavioural and emotional consequences of events, but also the underlying beliefs. This has made possible the connection with the Satir model that specifies facial and body expressions, voice intonation and linguistic structures related to five universal communication patterns. The consideration of the communication styles makes it possible to link the expressive capabilities of the agents with the BDI cognitive processing and to manage them an integrated way.
Download

Paper Nr: 251
Title:

Hints of Independence in a Pre-scripted World: On Controlled Usage of Open-domain Language Models for Chatbots in Highly Sensitive Domains

Authors:

Erkan Başar, Iris Hendrickx, Emiel Krahmer, Gert-Jan de Bruijn and Tibor Bosse

Abstract: Open-domain large language models have progressed to generating natural-sounding and coherent text. Even though the generated texts appear human-like, the main stumbling block is that their output is never fully predictable, which runs the risk of resulting in harmful content such as false statements or inflammatory language. This makes it difficult to apply these models in highly sensitive domains including personal health counselling. Hence, most of the chatbots for highly sensitive domains are developed using pre-scripted approaches. Although pre-scripted approaches are highly controlled, they suffer from repetitiveness and scalability issues. In this paper, we explore the possibility of combining the best of both worlds. We propose and describe in detail a new, flexible expert-driven hybrid architecture for harnessing the benefits of large language models in a controlled manner for highly sensitive domains and discuss the expectations and challenges.
Download

Paper Nr: 287
Title:

Multiagent Resource Planning System for Utility Provision

Authors:

Sergei Kozhevnikov, Petr Skobelev and Miroslav Svítek

Abstract: The intensive economic and social development of smart cities faces the constraints of stable utility provision. This paper presents a multiagent client-centric smart grid management system for integrated gas, heat and electricity networks. The system is based on the new approach of agent’s negotiation implementing the strategies of straight and reverse recursion planning. It can be used as part of the Smart grid and Micro grid concepts to reduce the price for consumers and decrease the negative impact of peak loads for the suppliers. This approach corresponds to the fundamental principles of modern complexity theory, which uses the fundamental principles of self-organization and evolution inherent in the natural world.
Download