ICAART 2021 Abstracts

Area 1 - Artificial Intelligence

Full Papers

Paper Nr:	14
Title:	Variation-resistant Q-learning: Controlling and Utilizing Estimation Bias in Reinforcement Learning for Better Performance
Authors:	Andreas Pentaliotis and Marco Wiering
Abstract:	Q-learning is a reinforcement learning algorithm that has overestimation bias, because it learns the optimal action values by using a target that maximizes over uncertain action-value estimates. Although the overestimation bias of Q-learning is generally considered harmful, a recent study suggests that it could be either harmful or helpful depending on the reinforcement learning problem. In this paper, we propose a new Q-learning variant, called Variation-resistant Q-learning, to control and utilize estimation bias for better performance. Firstly, we present the tabular version of the algorithm and mathematically prove its convergence. Secondly, we combine the algorithm with function approximation. Finally, we present empirical results from three different experiments, in which we compared the performance of Variation-resistant Q-learning, Q-learning, and Double Q-learning. The empirical results show that Variation-resistant Q-learning can control and utilize estimation bias for better performance in the experimental tasks.
Download

Paper Nr:	19
Title:	Comparing Privacy Policies of Government Agencies and Companies: A Study using Machine-learning-based Privacy Policy Analysis Tools
Authors:	Razieh N. Zaeem and K. S. Barber
Abstract:	Companies and government agencies are motivated by different missions when collecting and using Personally Identifiable Information (PII). Companies have strong incentives to monetize such information, whereas government agencies are generally not-for-profit. Besides this difference in missions, they are subject to distinct regulations that govern their collection and use of PII. Yet, do privacy policies of companies and government agencies reflect these differences and distinctions? In this paper, we take advantage of two of the most recent machine-learning-based privacy policy analysis tools, Polisis and PrivacyCheck, and five corpora of over 800 privacy policies to answer this question. We discover that government agencies are considerably better in protecting (or not collecting for that matter) sensitive financial information, Social Security Numbers, and user location. On the other hand, many of them fail to directly address children’s privacy or describe security measures taken to protect user data. Furthermore, we observe that E.U government agencies perform well, with respect to notifying users of policy change, giving users the right to edit/delete their data, and limiting data retention. Our work confirms the common-sense understanding that government agencies collect less personal information than companies, but discovers nuances, as listed above, along the way. Finally, we make our data publicly available, enhancing reproducibility and enabling future analyses.
Download

Paper Nr:	22
Title:	Multi-scale Convolutional Neural Networks for the Prediction of Human-virus Protein Interactions
Authors:	Xiaodi Yang, Ziding Zhang and Stefan Wuchty
Abstract:	Allowing the prediction of human-virus protein-protein interactions (PPI), our algorithm is based on a Siamese Convolutional Neural Network architecture (CNN), accounting for pre-acquired protein evolutionary profiles (i.e. PSSM) as input. In combinations with a multilayer perceptron, we evaluate our model on a variety of human-virus PPI datasets and compare its results with traditional machine learning frameworks, a deep learning architecture and several other human-virus PPI prediction methods, showing superior performance. Furthermore, we propose two transfer learning methods, allowing the reliable prediction of interactions in cross-viral settings, where we train our system with PPIs in a source human-virus domain and predict interactions in a target human-virus domain. Notable, we observed that our transfer learning approaches allowed the reliable prediction of PPIs in relatively less investigated human-virus domains, such as Dengue, Zika and SARS-CoV-2.
Download

Paper Nr:	23
Title:	Acoustic Anomaly Detection for Machine Sounds based on Image Transfer Learning
Authors:	Robert Müller, Fabian Ritz, Steffen Illium and Claudia Linnhoff-Popien
Abstract:	In industrial applications, the early detection of malfunctioning factory machinery is crucial. In this paper, we consider acoustic malfunction detection via transfer learning. Contrary to the majority of current approaches which are based on deep autoencoders, we propose to extract features using neural networks that were pre-trained on the task of image classification. We then use these features to train a variety of anomaly detection models and show that this improves results compared to convolutional autoencoders in recordings of four different factory machines in noisy environments. Moreover, we find that features extracted from ResNet based networks yield better results than those from AlexNet and Squeezenet. In our setting, Gaussian Mixture Models and One-Class Support Vector Machines achieve the best anomaly detection performance.
Download

Paper Nr:	32
Title:	Robustness-driven Exploration with Probabilistic Metric Temporal Logic
Authors:	Xiaotian Liu, Pengyi Shi, Tongtong Liu, Sarra Alqahtani, Paul Pauca and Miles Silman
Abstract:	The ability to perform autonomous exploration is essential for unmanned aerial vehicles (UAV) operating in unknown environments where it is difficult to describe the environment beforehand. Algorithms for autonomous exploration often focus on optimizing time and full coverage in a greedy fashion. These algorithms can collect irrelevant data and wastes time navigating areas with no important information. In this paper, we aim to improve the efficiency of exploration by maximizing the probability of detecting valuable information. The proposed approach relies on a theory of robustness based on Probabilistic Metric Temporal Logic (P-MTL) which is traditionally applied to offline verification and online control of hybrid systems. The robustness values would guide the UAV towards areas with more significant information by maximizing the satisfaction of the predefined P-MTL specifications. Markov Chain Monte Carlo (MCMC) is utilized to solve the P-MTL constraints. We tested our approach over Amazonian rainforest to detect areas occupied by illegal Artisanal Small-scale Gold Mining (ASGM) activities. The results show that our approach outperforms a greedy exploration approach from the literature by 38% in terms of ASGM coverage.
Download

Paper Nr:	38
Title:	Causal Campbell-Goodhart’s Law and Reinforcement Learning
Authors:	Hal Ashton
Abstract:	Campbell-Goodhart’s law relates to the causal inference error whereby decision-making agents aim to influence variables which are correlated to their goal objective but do not reliably cause it. This is a well known error in Economics and Political Science but not widely labelled in Artificial Intelligence research. Through a simple example, we show how off-the-shelf deep Reinforcement Learning (RL) algorithms are not necessarily immune to this cognitive error. The off-policy learning method is tricked, whilst the on-policy method is not. The practical implication is that naive application of RL to complex real life problems can result in the same types of policy errors that humans make. Great care should be taken around understanding the causal model that underpins a solution derived from Reinforcement Learning.
Download

Paper Nr:	57
Title:	Online Learning of non-Markovian Reward Models
Authors:	Gavin Rens, Jean-François Raskin, Raphaël Reynouard and Giuseppe Marra
Abstract:	There are situations in which an agent should receive rewards only after having accomplished a series of previous tasks, that is, rewards are non-Markovian. One natural and quite general way to represent history- dependent rewards is via a Mealy machine. In our formal setting, we consider a Markov decision process (MDP) that models the dynamics of the environment in which the agent evolves and a Mealy machine synchronized with this MDP to formalize the non-Markovian reward function. While the MDP is known by the agent, the reward function is unknown to the agent and must be learned. Our approach to overcome this challenge is to use Angluin’s L∗ active learning algorithm to learn a Mealy machine representing the underlying non-Markovian reward machine (MRM). Formal methods are used to determine the optimal strategy for answering so-called membership queries posed by L∗. Moreover, we prove that the expected reward achieved will eventually be at least as much as a given, reasonable value provided by a domain expert. We evaluate our framework on two problems. The results show that using L∗ to learn an MRM in a non-Markovian reward decision process is effective.
Download

Paper Nr:	63
Title:	Twin-GAN for Neural Machine Translation
Authors:	Jiaxu Zhao, Li Huang, Ruixuan Sun, Liao Bing and Hong Qu
Abstract:	In recent years, Neural Machine Translation (NMT) has achieved great success, but we can not ignore two important problems. One is the exposure bias caused by the different strategies between training and inference, and the other is that the NMT model generates the best candidate word for the current step yet a bad element of the whole sentence. The popular methods to solve these two problems are Schedule Sampling and Generative Adversarial Networks (GANs) respectively, and both achieved some success. In this paper, we proposed a more precise approach called “similarity selection” combining a new GAN structure called twin-GAN to solve the above two problems. There are two generators and two discriminators in the twin-GAN. One generator uses the “similarity selection” and the other one uses the same way as inference (simulate the inference process). One discriminator guides generators at the sentence level, and the other discriminator forces these two generators to have similar distributions. Moreover, we performed a lot of experiments on the IWSLT 2014 German→English (De→En) and the WMT’17 Chinese!English (Zh→En) and the result shows that we improved the performance compared to some other strong baseline models which based on recurrent architecture.
Download

Paper Nr:	74
Title:	Analysis of Feature Representations for Anomalous Sound Detection
Authors:	Robert Müller, Steffen Illium, Fabian Ritz and Kyrill Schmid
Abstract:	In this work, we thoroughly evaluate the efficacy of pretrained neural networks as feature extractors for anomalous sound detection. In doing so, we leverage the knowledge that is contained in these neural networks to extract semantically rich features (representations) that serve as input to a Gaussian Mixture Model which is used as a density estimator to model normality. We compare feature extractors that were trained on data from various domains, namely: images, environmental sounds and music. Our approach is evaluated on recordings from factory machinery such as valves, pumps, sliders and fans. All of the evaluated representations outperform the autoencoder baseline with music based representations yielding the best performance in most cases. These results challenge the common assumption that closely matching the domain of the feature extractor and the downstream task results in better downstream task performance.
Download

Paper Nr:	76
Title:	An Investigation Into the Effect of the Learning Rate on Overestimation Bias of Connectionist Q-learning
Authors:	Yifei Chen, Lambert Schomaker and Marco Wiering
Abstract:	In Reinforcement learning, Q-learning is the best-known algorithm but it suffers from overestimation bias, which may lead to poor performance or unstable learning. In this paper, we present a novel analysis of this problem using various control tasks. For solving these tasks, Q-learning is combined with a multilayer perceptron (MLP), experience replay, and a target network. We focus our analysis on the effect of the learning rate when training the MLP. Furthermore, we examine if decaying the learning rate over time has advantages over static ones. Experiments have been performed using various maze-solving problems involving deterministic or stochastic transition functions and 2D or 3D grids and two Open-AI gym control problems. We conducted the same experiments with Double Q-learning using two MLPs with the same parameter settings, but without target networks. The results on the maze problems show that for Q-learning combined with the MLP, the overestimation occurs when higher learning rates are used and not when lower learning rates are used. The Double Q-learning variant becomes much less stable with higher learning rates and with low learning rates the overestimation bias may still occur. Overall, decaying learning rates clearly improves the performances of both Q-learning and Double Q-learning.
Download

Paper Nr:	83
Title:	Discrete and Continuous Deep Residual Learning over Graphs
Authors:	Pedro C. Avelar, Anderson R. Tavares, Marco Gori and Luís C. Lamb
Abstract:	We propose the use of continuous residual modules for graph kernels in Graph Neural Networks. We show how both discrete and continuous residual layers allow for more robust training, being that continuous residual layers are applied by integrating through an Ordinary Differential Equation (ODE) solver to produce their output. We experimentally show that these modules achieve better results than the ones with non-residual modules when multiple layers are used, thus mitigating the low-pass filtering effect of Graph Convolutional Network-based models. Finally, we discuss the behaviour of discrete and continuous residual layers, pointing out possible domains where they could be useful by allowing more predictable behaviour under dynamic times of computation.
Download

Paper Nr:	86
Title:	A Framework of Hierarchical Deep Q-Network for Portfolio Management
Authors:	Yuan Gao, Ziming Gao, Yi Hu, Sifan Song, Zhengyong Jiang and Jionglong Su
Abstract:	Reinforcement Learning algorithms and Neural Networks have diverse applications in many domains, e.g., stock market prediction, facial recognition and automatic machine translation. The concept of modeling the portfolio management through a reinforcement learning formulation is novel, and the Deep Q-Network has been successfully applied to portfolio management recently. However, the model does not take into account of commission fee for transaction. This paper introduces a framework, based on the hierarchical Deep Q-Network, that addresses the issue of zero commission fee by reducing the number of assets assigned to each Deep Q-Network and dividing the total portfolio value into smaller parts. Furthermore, this framework is flexible enough to handle an arbitrary number of assets. In our experiments, the time series of four stocks for three different time periods are used to assess the efficacy of our model. It is found that our hierarchical Deep Q-Network based strategy outperforms ten other strategies, including nine traditional strategies and one reinforcement learning strategy, in profitability as measured by the Cumulative Rate of Return. Moreover, the Sharpe ratio and Max Drawdown metrics both demonstrate that the risk of policy associated with hierarchical Deep Q-Network is the lowest among all ten strategies.
Download

Paper Nr:	88
Title:	Sensor Fusion Neural Networks for Gesture Recognition on Low-power Edge Devices
Authors:	Gabor Balazs, Mateusz Chmurski, Walter Stechele and Mariusz Zubert
Abstract:	The goal of hand gesture recognition based on time-of-flight and radar sensors is to enhance the human-machine interface, while taking care of privacy issues of camera sensors. Additionally, the system needs to be deployable on low-power edge devices for applicability in serial-produced vehicles. Recent advances show the capabilities of deep neural networks for gesture classification but they are still limited to high performance hardware. Embedded neural network accelerators are constrained in memory and supported operations. These limitations form an architectural design problem that is addressed in this work. Novel gesture classification networks are optimized for embedded deployment. The new approaches perform equally compared to high-performance neural networks with 3D convolutions, but need only 8.9% of the memory. These lightweight network architectures allow deployment on constrained embedded accelerator devices, thus enhancing human-machine interfaces.
Download

Paper Nr:	89
Title:	Micro-YOLO: Exploring Efficient Methods to Compress CNN based Object Detection Model
Authors:	Lining Hu and Yongfu Li
Abstract:	Deep learning models have made significant breakthroughs in the performance of object detection. However, in the traditional models, such as Faster R-CNN and YOLO, the size of these networks make it too difficult to be deployed on embedded mobile devices due to limited computation resources and tight power budgets. Hence, we propose a new light-weight CNN based object detection model, Micro-YOLO based on YOLOv3-Tiny, which achieves a signification reduction in the number of parameters and computation cost while maintaining the detection performance. We propose to replace convolutional layers in the YOLOv3-tiny network with the Depth-wise Separable convolution (DSConv) and the mobile inverted bottleneck convolution with squeeze and excitation block (MBConv), and design a progressive channel-level pruning algorithm to minimize the number of parameters and maximize the detection performance. Hence, the proposed MicroYOLO network reduces the number of parameters by 3.46× and multiply-accumulate operation (MAC) by 2.55× while slightly decreases the mAP evaluated on the COCO dataset by 0.7%, compared to the original YOLOv3-tiny network.
Download

Paper Nr:	91
Title:	ImpactCite: An XLNet-based Solution Enabling Qualitative Citation Impact Analysis Utilizing Sentiment and Intent
Authors:	Dominique Mercier, Syed R. Rizvi, Vikas Rajashekar, Andreas Dengel and Sheraz Ahmed
Abstract:	Citations play a vital role in understanding the impact of scientific literature. Generally, citations are analyzed quantitatively whereas qualitative analysis of citations can reveal deeper insights into the impact of a scientific artifact in the community. Therefore, citation impact analysis including sentiment and intent classification enables us to quantify the quality of the citations which can eventually assist us in the estimation of ranking and impact. The contribution of this paper is three-fold. First, we provide ImpactCite, which is an XLNet-based method for citation impact analysis. Second, we propose a clean and reliable dataset for citation sentiment analysis. Third, we benchmark the well-known language models like BERT and ALBERT along with our proposed approach for both tasks of sentiment and intent classification. All evaluations are performed on a set of publicly available citation analysis datasets. Evaluation results reveal that ImpactCite achieves a new state-of-the-art performance for both citation intent and sentiment classification by outperforming the existing approaches by 3.44% and 1.33% in F1-score. Therefore, the evaluation results suggest that ImpactCite is a single solution for both sentiment and intent analysis to better understand the impact of a citation.
Download

Paper Nr:	97
Title:	Compiling Possibilistic Networks to Compute Learning Indicators
Authors:	Guillaume Petiot
Abstract:	University teachers, who generally focus their interest on pedagogy and students, may find it difficult to manage e-learning platforms which provide learning analytics and data. But learning indicators might help teachers when the amount of information to process grows exponentially. The indicators can be computed by the aggregation of data and by using teachers’ knowledge which is often imprecise and uncertain. Possibility theory provides a solution to handle these drawbacks. Possibilistic networks allow us to represent the causal link between the data but they require the definition of all the parameters of Conditional Possibility Tables. Uncertain gates allow the automatic calculation of these Conditional Possibility Tables by using for example the logical combination of information. The calculation time to propagate new evidence in possibilistic networks can be improved by compiling possibilistic networks. In this paper, we will present an experimentation of compiling possibilistic networks to compute course indicators. Indeed, the LMS Moodle provides a large scale of data about learners that can be merged to provide indicators to teachers in a decision making system. Thus, teachers can propose differentiated instruction which, better corresponds to their student’s expectations and their learning style.
Download

Paper Nr:	101
Title:	YOdar: Uncertainty-based Sensor Fusion for Vehicle Detection with Camera and Radar Sensors
Authors:	Kamil Kowol, Matthias Rottmann, Stefan Bracke and Hanno Gottschalk
Abstract:	In this work, we present an uncertainty-based method for sensor fusion with camera and radar data. The outputs of two neural networks, one processing camera and the other one radar data, are combined in an uncertainty aware manner. To this end, we gather the outputs and corresponding meta information for both networks. For each predicted object, the gathered information is post-processed by a gradient boosting method to produce a joint prediction of both networks. In our experiments we combine the YOLOv3 object detection network with a customized 1D radar segmentation network and evaluate our method on the nuScenes dataset. In particular we focus on night scenes, where the capability of object detection networks based on camera data is potentially handicapped. Our experiments show, that this approach of uncertainty aware fusion, which is also of very modular nature, significantly gains performance compared to single sensor baselines and is in range of specifically tailored deep learning based fusion approaches.
Download

Paper Nr:	113
Title:	Can We Detect Harmony in Artistic Compositions? A Machine Learning Approach
Authors:	Adam Vandor, Marie Van Vollenhoven, Gerhard Weiss and Gerasimos Spanakis
Abstract:	Harmony in visual compositions is a concept that cannot be defined or easily expressed mathematically, even by humans. The goal of the research described in this paper was to find a numerical representation of artistic compositions with different levels of harmony. We ask humans to rate a collection of grayscale images based on the harmony they convey. To represent the images, a set of special features were designed and extracted. By doing so, it became possible to assign objective measures to subjectively judged compositions. Given the ratings and the extracted features, we utilized machine learning algorithms to evaluate the efficiency of such representations in a harmony classification problem. The best performing model (SVM) achieved 80% accuracy in distinguishing between harmonic and disharmonic images, which reinforces the assumption that concept of harmony can be expressed in a mathematical way that can be assessed by humans.
Download

Paper Nr:	119
Title:	A New Parking Space Allocation System based on a Distributed Constraint Optimization Approach
Authors:	Atik Ali, Souhila Arib and Samir Aknine
Abstract:	This paper develops and evaluates a new decentralized mechanism for the allocation of parking slots in downtown, using a distributed constraints optimization approach (DCOP). Our mechanism works with the multi- parking/multi-zone model, where vehicles are connected and can exchange information with the distributed allocation system. This mechanism can reach the minimal allocation costs where vehicles are assigned to the parking lots with the best possible aggregated user costs. The cost is calculated based on driver’s aggregated preferences over slots. We empirically evaluated the performance of our approach with randomly generated costs and tested on three different configurations. The evaluation shows the performance of each configuration in terms of runtime and volume of exchanged data.
Download

Paper Nr:	121
Title:	Beneficial Effect of Combined Replay for Continual Learning
Authors:	M. Solinas, S. Rousset, R. Cohendet, Y. Bourrier, M. Mainsant, A. Molnos, M. Reyboz and M. Mermillod
Abstract:	While deep learning has yielded remarkable results in a wide range of applications, artificial neural networks suffer from catastrophic forgetting of old knowledge as new knowledge is learned. Rehearsal methods overcome catastrophic forgetting by replaying an amount of previously learned data stored in dedicated memory buffers. Alternatively, pseudo-rehearsal methods generate pseudo-samples to emulate the previously learned data, thus alleviating the need for dedicated buffers. Unfortunately, up to now, these methods have shown limited accuracy. In this work, we combine these two approaches and employ the data stored in tiny memory buffers as seeds to enhance the pseudo-sample generation process. We then show that pseudo-rehearsal can improve performance versus rehearsal methods for small buffer sizes. This is due to an improvement in the retrieval process of previously learned information. Our combined replay approach consists of a hybrid architecture that generates pseudo-samples through a reinjection sampling procedure (i.e. iterative sampling). The generated pseudo-samples are then interlaced with the new data to acquire new knowledge without forgetting the previous one. We evaluate our method extensively on the MNIST, CIFAR-10 and CIFAR-100 image classification datasets, and present state-of-the-art performance using tiny memory buffers.
Download

Paper Nr:	129
Title:	Portable High-level Agent Programming with golog++
Authors:	Victor Mataré, Tarik Viehmann, Till Hofmann, Gerhard Lakemeyer, Alexander Ferrein and Stefan Schiffer
Abstract:	We present golog++, a high-level agent programming and interfacing framework that offers a temporal constraint language to explicitly model layer-penetrating contingencies in low-level platform behavior. It can be used to maintain a clear separation between an agent’s domain model and certain quirks of its execution platform that affect problem solving behavior. Our system reasons about the execution of an abstract (i.e. exclusively domain-bound) plan on a particular execution platform. This way, we avoid compounding the complexity of the planning problem while improving the modularity of both golog++ and the user code. On a run-through example from the well-known blocksworld domain, we demonstrate the entire process from domain modeling and platform modeling to plan transformation and platform-specific plan execution.
Download

Paper Nr:	135
Title:	Explaining Inaccurate Predictions of Models through k-Nearest Neighbors
Authors:	Zeki Bilgin and Murat Gunestas
Abstract:	Deep Learning (DL) models exhibit dramatic success in a wide variety of fields such as human-machine interaction, computer vision, speech recognition, etc. Yet, the widespread deployment of these models partly depends on earning trust in them. Understanding how DL models reach a decision can help to build trust on these systems. In this study, we present a method for explaining inaccurate predictions of DL models through post-hoc analysis of k-nearest neighbours. More specifically, we extract k-nearest neighbours from training samples for a given mispredicted test instance, and then feed them into the model as input to observe the model’s response which is used for post-hoc analysis in comparison with the original mispredicted test sample. We apply our method on two different datasets, i.e. IRIS and CIFAR10, to show its feasibility on concrete examples.
Download

Paper Nr:	137
Title:	Reinforcement Learning with Quantitative Verification for Assured Multi-Agent Policies
Authors:	Joshua Riley, Radu Calinescu, Colin Paterson, Daniel Kudenko and Alec Banks
Abstract:	In multi-agent reinforcement learning, several agents converge together towards optimal policies that solve complex decision-making problems. This convergence process is inherently stochastic, meaning that its use in safety-critical domains can be problematic. To address this issue, we introduce a new approach that combines multi-agent reinforcement learning with a formal verification technique termed quantitative verification. Our assured multi-agent reinforcement learning approach constrains agent behaviours in ways that ensure the satisfaction of requirements associated with the safety, reliability, and other non-functional aspects of the decision-making problem being solved. The approach comprises three stages. First, it models the problem as an abstract Markov decision process, allowing quantitative verification to be applied. Next, this abstract model is used to synthesise a policy which satisfies safety, reliability, and performance constraints. Finally, the synthesised policy is used to constrain agent behaviour within the low-level problem with a greatly lowered risk of constraint violations. We demonstrate our approach using a safety-critical multi-agent patrolling problem.
Download

Paper Nr:	143
Title:	HNAS: Hyper Neural Architecture Search for Image Segmentation
Authors:	Yassir Houreh, Mahsa Mahdinejad, Enrique Naredo, Douglas M. Dias and Conor Ryan
Abstract:	Deep learning is a well suited approach to successfully address image processing and there are several Neural Networks architectures proposed on this research field, one interesting example is the U-net architecture and and its variants. This work proposes to automatically find the best architecture combination from a set of the current most relevant U-net architectures by using a genetic algorithm (GA) applied to solve the Retinal Blood Vessel Segmentation (RVS), which it is relevant to diagnose and cure blindness in diabetes patients. Interestingly, the experimental results show that avoiding human-bias in the design, GA finds novel combinations of U-net architectures, which at first sight seems to be complex but it turns out to be smaller, reaching competitive performance than the manually designed architectures and reducing considerably the computational effort to evolve them.
Download

Paper Nr:	147
Title:	WICO Graph: A Labeled Dataset of Twitter Subgraphs based on Conspiracy Theory and 5G-Corona Misinformation Tweets
Authors:	Daniel T. Schroeder, Ferdinand Schaal, Petra Filkukova, Konstantin Pogorelov and Johannes Langguth
Abstract:	In the wake of the COVID-19 pandemic, a surge of misinformation has flooded social media and other internet channels, and some of it has the potential to cause real-world harm. To counteract this misinformation, reliably identifying it is a principal problem to be solved. However, the identification of misinformation poses a formidable challenge for language processing systems since the texts containing misinformation are short, work with insinuation rather than explicitly stating a false claim, or resemble other postings that deal with the same topic ironically. Accordingly, for the development of better detection systems, it is not only essential to use hand-labeled ground truth data and extend the analysis with methods beyond Natural Language Processing to consider the characteristics of the participant’s relationships and the diffusion of misinformation. This paper presents a novel dataset that deals with a specific piece of misinformation: the idea that the 5G wireless network is causally connected to the COVID-19 pandemic. We have extracted the subgraphs of 3,000 manually classified Tweets from Twitter’s follower network and distinguished them into three categories. First, subgraphs of Tweets that propagate the specific 5G misinformation, those that spread other conspiracy theories, and Tweets that do neither. We created the WICO (Wireless Networks and Coronavirus Conspiracy) dataset to support experts in machine learning experts, graph processing, and related fields in studying the spread of misinformation. Furthermore, we provide a series of baseline experiments using both Graph Neural Networks and other established classifiers that use simple graph metrics as features. The dataset is available at https://datasets.simula.no/wico-graph..
Download

Paper Nr:	159
Title:	A Deep Q-learning based Path Planning and Navigation System for Firefighting Environments
Authors:	Manish Bhattarai and Manel Martínez-Ramón
Abstract:	Live fire creates a dynamic, rapidly changing environment that presents a worthy challenge for deep learning and artificial intelligence methodologies to assist firefighters with scene comprehension in maintaining their situational awareness, tracking and relay of important features necessary for key decisions as they tackle these catastrophic events. We propose a deep Q-learning based agent who is immune to stress induced disorientation and anxiety and thus able to make clear decisions for firefighter navigation based on the observed and stored facts in live fire environments. As a proof of concept, we imitate structural fire in a gaming engine called Unreal Engine which enables the interaction of the agent with the environment. The agent is trained with a deep Q-learning algorithm based on a set of rewards and penalties as per its actions on the environment. We exploit experience replay to accelerate the learning process and augment the learning of the agent with human-derived experiences. The agent trained under this deep Q-learning approach outperforms agents trained through alternative path planning systems and demonstrates this methodology as a promising foundation on which to build a path planning navigation assistant. This assistant is capable of safely guiding firefighters through live-fire environments in fireground navigation activities that range from exploration to personnel rescue.
Download

Paper Nr:	162
Title:	Dimensionality Reduction and Bandwidth Selection for Spatial Kernel Discriminant Analysis
Authors:	Soumia Boumeddane, Leila Hamdad, Hamid Haddadou and Sophie Dabo-Niang
Abstract:	Spatial Kernel Discriminant Analysis is a powerful tool for the classification of spatially dependent data. It allows taking into consideration the spatial autocorrelation of data based on a spatial kernel density estimator. The performance of SKDA is highly influenced by the choice of the smoothing parameters, also known as bandwidths. Moreover, computing a kernel density estimate is computationally intensive for high-dimensional datasets. In this paper, we consider the bandwidth selection as an optimization problem, that we resolve using Particle Swarm Optimization algorithm. In addition, we investigate the use of Principle Component Analysis as a feature extraction technique to reduce computational complexity and overcome curse of dimensionality drawback. We examined the performance of our model on Hyperspectral image classification. Experiments have given promising results on a commonly used dataset.
Download

Paper Nr:	163
Title:	Web-Gorgias-B: Argumentation for All
Authors:	Nikolaos I. Spanoudakis, Konstantinos Kostis and Katerina Mania
Abstract:	This paper proposes the use of a web-based authoring tool for the development of applications of argumentation. It focuses on aiding people that have little, or no knowledge of logic programming, or of an argumentation framework, to develop argumentation-based decision policies. To achieve this, it proposes an implementation of the table formalism that has recently been proposed in the literature. The proposed implementation contains original features that were evaluated by experts in web-application development, students and experts in argumentation. The main feature of the proposed system is the ability to define a default preferred option in a given scenario, thus, allowing for other options to be used in further refinements of the scenario. We followed a user-centered development process using the think aloud protocol. We evaluated the usability of the system with the System Usability Scale, validating our hypothesis that even naive users can employ it to define their decision policies.
Download

Paper Nr:	164
Title:	VICA: A Vicarious Cognitive Architecture Environment Model for Navigation Among Movable Obstacles
Authors:	Halim Djerroud and Arab Ali-Chérif
Abstract:	This article presents a new Cognitive Architecture Environment model for Navigation Among Movable Obstacles (NAMO). This model is the result of a novel approach based on the Theory of Mind and more particularly on the notion of ’vicariance’ as an essential strategy of the robot’s interaction with outside world. The implementation of our model follows the advances in AI and the Cognitive Robotics research area, where a cognitive architecture environment is represented as a Multi-Agent System (MAS). The MAS representation offers the robot the ability to produce a representation of its environment as well as the possibility to run all types of action simulations in order to anticipate the environment’s reactions. The environment state values, both predictive and real as transcribed during simulation and real action movements, are compared to each other in order to keep the correct ones and avoid errors. This is a continuous learning and leads to the construction of a safe path of actions into a dynamic environment. The experiment results show the efficiency of our model, offering an intelligent guide to the robot in order to perform tasks among mobile agents, by avoiding a maximum number of obstacles.
Download

Paper Nr:	171
Title:	Acoustic Leak Detection in Water Networks
Authors:	Robert Müller, Steffen Illium, Fabian Ritz, Tobias Schröder, Christian Platschek, Jörg Ochs and Claudia Linnhoff-Popien
Abstract:	In this work, we present a general procedure for acoustic leak detection in water networks that satisfies multiple real-world constraints such as energy efficiency and ease of deployment. Based on recordings from seven contact microphones attached to the water supply network of a municipal suburb, we trained several shallow and deep anomaly detection models. Inspired by how human experts detect leaks using electronic sounding-sticks, we use these models to repeatedly listen for leaks over a predefined decision horizon. This way we avoid constant monitoring of the system. While we found the detection of leaks in close proximity to be a trivial task for almost all models, neural network based approaches achieve better results at the detection of distant leaks.
Download

Paper Nr:	202
Title:	ChronSeg: Novel Dataset for Segmentation of Handwritten Historical Chronicles
Authors:	Josef Baloun, Pavel Král and Ladislav Lenc
Abstract:	The segmentation of document images plays an important role in the process of making their content electronically accessible. This work focuses on the segmentation of historical handwritten documents, namely chronicles. We take image, text and background classes into account. For this goal, a new dataset is created mainly from chronicles provided by Porta fontium. In total, the dataset consists of 58 images of document pages and their precise annotations for text, image and graphic regions in PAGE format. The annotations are also provided at a pixel level. Further, we present a baseline evaluation using an approach based on a fully convolutional neural network. We also perform a series of experiments in order to identify the best method configuration. It includes a novel data augmentation method which creates artificial pages.
Download

Paper Nr:	205
Title:	On Explanation of Propositional Logic-based Argumentation System
Authors:	Teeradaj Racharak and Satoshi Tojo
Abstract:	We present a characterization about argumentation and proof in logic. Indeed, we show that proof for a claim α from a set of premises Φ can be deemed as a structured form of an argument for that claim. Due to the expressivity of classical propositional logic (PL), this work considers that the knowledge-base is represented in PL, in which the semantics and proof systems for individual arguments are studied and utilized. We show that natural deduction (ND) can be used as a basis of proof for an argument and also for modeling counterarguments in the form of canonical undercut. We reveal that ND does not merely enable for the construction of arguments but also paves the way naturally for a human-understandable form of argumentative reasoning. Finally, we demonstrate that our approach gives the feasibility of developing explainable artificial intelligence systems that can offer human-friendly explanations to the users.
Download

Paper Nr:	206
Title:	Construct-Extract: An Effective Model for Building Bilingual Corpus to Improve English-Myanmar Machine Translation
Authors:	May M. Zin, Teeradaj Racharak and Nguyen M. Le
Abstract:	When dealing with low resource languages such as Myanmar, using additional pseudo parallel data for training machine translation systems is often an effective approach. As a pseudo parallel corpus is generated by back-translating target monolingual texts into the source language, it potentially contains a lot of noise including translation errors and weakly paired sentences and is thus required cleaning. In this paper, we propose a noisy parallel-sentences filtering system called Construct-Extract based on cosine similarity and Siamese BERT-Networks based cross-lingual sentence embeddings. The proposed system filters out noisy sentences by extracting high score sentence pairs from the constructed pseudo parallel data to finally obtain better synthetic parallel data. As part of the proposed system, we also introduce an unsupervised Myanmar sub-word segmenter to improve the quality of current English-Myanmar translation models that are potential to be used as backward systems for back-translation and often suffer from Myanmar word segmentation errors. Experiments show that the proposed Myanmar word segmentation could help the backward system to construct more accurate back-translated pseudo parallel data and using our extracted pseudo parallel corpus led to improve the performance of English-Myanmar translation systems in the two directions.
Download

Paper Nr:	216
Title:	Mathematical Programming Approach for Adversarial Attack Modelling
Authors:	Hatem Ibn-Khedher, Mohamed Ibn Khedher and Makhlouf Hadji
Abstract:	An adversarial attack is defined as the minimal perturbation that change the model decision. Machine learning (ML) models such as Deep Neural Networks (DNNs) are vulnerable to different adversarial examples where malicious perturbed inputs lead to erroneous model outputs. Breaking neural networks with adversarial attack requires an intelligent approach that decides about the maximum allowed margin in which the neural network decision (output) is invariant. In this paper, we propose a new formulation based on linear programming approach modelling adversarial attacks. Our approach considers noised inputs while reaching the optimal perturbation. To assess the performance of our approach, we discuss two main scenarios quantifying the algorithm’s decision behavior in terms of total perturbation cost, percentage of perturbed inputs, and other cost factors. Then, the approach is implemented and evaluated under different neural network scales.
Download

Paper Nr:	217
Title:	TWIN-GRU: Twin Stream GRU Network for Action Recognition from RGB Video
Authors:	Hajer Essefi, Olfa Ben Ahmed, Christel Bidet-Ildei, Yannick Blandin and Christine Fernandez-Maloigne
Abstract:	Human Action Recognition (HAR) is an important task for numerous computer vision applications. Recently, deep learning approaches have shown proficiency in recognizing actions in RGB video. However, existing models rely mainly on global appearance and could potentially under perform in real world applications, such as sport events and clinical applications. Refereeing to domain knowledge in how human perceive action, we hypothesis that observing the dynamic of a 2D human body joints representation extracted from RGB video frames is sufficient to recognize an action in video. Moreover, body joints contain structural information with a strong spatial (intra-frame) and temporal (inter-frame) correlation between adjacent joints. In this paper, we propose a psychology-inspired twin stream Gated Recurrent Unit network for action recognition based on the dynamic of 2D human body joints in RGB videos. The proposed model achieves a classification accuracy of 89,97% in a subject-specific experiment and outperforms the baseline method that fuses depth and inertial sensor data on the UTD-MHAD dataset. The proposed framework is more cost effective and highly competitive than depth 3D skeleton based solutions and therefore can be used outside capture motion labs for real world applications.
Download

Paper Nr:	222
Title:	Secrecy-preserving Reasoning in Acyclic DL-LiteR Knowledge Bases in the Presence of BCQs
Authors:	Gopalakrishnan Krishnasamy-Sivaprakasam and Giora Slutzki
Abstract:	In this paper we study Secrecy-Preserving Query Answering under Open World Assumption (OWA) for DL-LiteR Knowledge Bases (KBs) with acyclic TBox. Using a tableau algorithm, we construct A∗, an inferential closure of the given ABox A, which includes both positive as well as negative assertions. We use a notational variant of Kleene 3-valued semantics, which we call OW-semantics as it fits nicely with OWA. This allows us to answer queries, including Boolean Conjunctive Queries (BCQs) with “Yes”, “No” or “Unknown”, as opposed to the just answering “Yes” or “No” as in Ontology Based Data Access (OBDA) framework, thus improving the informativeness of the query-answering procedure. Being able to answer “Unknown” plays a key role in protecting secrecy under OWA. One of the main contributions of this paper is a study of answering BCQs without compromising secrecy. Using the idea of secrecy envelopes, previously introduced by one of the authors, we give a precise characterization of when BCQs should be answered “Yes”, “No” or “Unknown”. We prove the correctness of the secrecy-preserving query-answering algorithm.
Download

Paper Nr:	238
Title:	A Novel Deep Learning Power Quality Disturbance Classification Method using Autoencoders
Authors:	Callum O’Donovan, Cinzia Giannetti and Grazia Todeschini
Abstract:	Automatic identification and classification of power quality disturbances (PQDs) is crucial for maintaining efficiency and safety of electrical systems and equipment condition. In recent years emerging deep learning techniques have shown potential in performing classification of PQDs. This paper proposes two novel deep learning models, called CNN(AE)-LSTM and CNN-LSTM(AE) that automatically distinguish between normal power system behaviour and three types of PQDs: voltage sags, voltage swells and interruptions. The CNN-LSTM(AE) model achieved the highest average classification accuracy with a 65:35 train-test split. The Adam optimiser and a learning rate of 0.001 were used for ten epochs with a batch size of 64. Both models are trained using real world data and outperform models found in literature. This work demonstrates the potential of deep learning in classifying PQDs and hence paves the way to effective implementation of AI-based automated quality monitoring to identify disturbances and reduce failures in real world power systems.
Download

Paper Nr:	250
Title:	Informer, an Information Organization Transformer Architecture
Authors:	Cristian E. Ojeda, Cayetano G. Artal and Francisco H. Tejera
Abstract:	The use of architectures based on transformers presents a state of the art revolution in natural language processing (NLP). The employment of these architectures with high computational costs has increased in the last few months, despite the existing use of parallelization techniques. This is due to the high performance that is obtained by increasing the size of the learnable parameters for these kinds of architectures, while maintaining the models’ predictability. This relates to the fact that it is difficult to do research with limited computational resources. A restrictive element is the memory usage, which seriously affects the replication of experiments. We are presenting a new architecture called Informer, which seeks to exploit the concept of information organization. For the sake of evaluation, we use a neural machine translation (NMT) dataset, the English-Vietnamese IWSLT15 dataset (Luong and Manning, 2015). In this paper, we also compare this proposal with architectures that reduce the computational cost to O(n · r), such as Linformer (Wang et al., 2020). In addition, we have managed to improve the SOTA of the BLEU score from 33.27 to 35.11.
Download

Paper Nr:	252
Title:	An Ensemble-based Approach by Fine-Tuning the Deep Transfer Learning Models to Classify Pneumonia from Chest X-Ray Images
Authors:	Sagar Kora Venu
Abstract:	Pneumonia is caused by viruses, bacteria, or fungi that infect the lungs, which, if not diagnosed and treated in time, can be fatal and lead to respiratory failure. More than 250,000 individuals in the United States, mainly adults, are diagnosed with pneumonia each year, and 50,000 die from the disease. Chest Radiography (X-ray) is widely used by radiologists to detect pneumonia. It is not uncommon to overlook pneumonia detection for a well-trained radiologist, which triggers the need for improvement in the accuracy of the diagnosis. Therefore, we propose using transfer learning, which can reduce the neural network’s training time and minimize the generalization error to improve the accuracy of the diagnosis. We trained, fine-tuned the state-of-the-art deep learning models such as InceptionResNet, MobileNetV2, Xception, DenseNet201, and ResNet152V2 to classify pneumonia accurately. Later, we created a weighted average ensemble of these models and achieved a test accuracy of 98.46%, precision of 98.38%, recall of 99.53%, and f1 score of 98.96%. These performance metrics of accuracy, precision, and f1 score are at their highest levels ever reported in the literature, which can be considered a benchmark for the accurate pneumonia classification.
Download

Paper Nr:	257
Title:	Interpretability in Word Sense Disambiguation using Tsetlin Machine
Authors:	Rohan K. Yadav, Lei Jiao, Ole-Christoffer Granmo and Morten Goodwin
Abstract:	Word Sense Disambiguation (WSD) is a longstanding unresolved task in Natural Language Processing. The challenge lies in the fact that words with the same spelling can have completely different senses, sometimes depending on subtle characteristics of the context. A weakness of the state-of-the-art supervised models, however, is that it can be difficult to interpret them, making it harder to check if they capture senses accurately or not. In this paper, we introduce a novel Tsetlin Machine (TM) based supervised model that distinguishes word senses by means of conjunctive clauses. The clauses are formulated based on contextual cues, represented in propositional logic. Our experiments on CoarseWSD-balanced dataset indicate that the learned word senses can be relatively effortlessly interpreted by analyzing the converged model of the TM. Additionally, the classification accuracy is higher than that of FastText-Base and similar to that of FastText-CommonCrawl.
Download

Paper Nr:	258
Title:	Measuring the Novelty of Natural Language Text using the Conjunctive Clauses of a Tsetlin Machine Text Classifier
Authors:	Bimal Bhattarai, Ole-Christoffer Granmo and Lei Jiao
Abstract:	Most supervised text classification approaches assume a closed world, counting on all classes being present in the data at training time. This assumption can lead to unpredictable behaviour during operation, whenever novel, previously unseen, classes appear. Although deep learning-based methods have recently been used for novelty detection, they are challenging to interpret due to their black-box nature. This paper addresses interpretable open-world text classification, where the trained classifier must deal with novel classes during operation. To this end, we extend the recently introduced Tsetlin machine (TM) with a novelty scoring mechanism. The mechanism uses the conjunctive clauses of the TM to measure to what degree a text matches the classes covered by the training data. We demonstrate that the clauses provide a succinct interpretable description of known topics, and that our scoring mechanism makes it possible to discern novel topics from the known ones. Empirically, our TM-based approach outperforms seven other novelty detection schemes on three out of five datasets, and performs second and third best on the remaining, with the added benefit of an interpretable propositional logic-based representation.
Download

Paper Nr:	264
Title:	LSTM-based System for Multiple Obstacle Detection using Ultra-wide Band Radar
Authors:	Amira Mimouna, Anouar Ben Khalifa, Ihsen Alouani, Abdelmalik Taleb-Ahmed, Atika Rivenq and Najoua E. Ben Amara
Abstract:	Autonomous vehicles present a promising opportunity in the future of transportation systems by providing road safety. As significant progress has been made in the automatic environment perception, the detection of road obstacles remains a major challenge. Thus, to achieve reliable obstacle detection, several sensors have been employed. For short ranges, the Ultra-Wide Band (UWB) radar is utilized in order to detect objects in the near field. However, the main challenge appears in distinguishing the real target’s signature from noise in the received UWB signals. In this paper, we propose a novel framework that exploits Recurrent Neural Networks (RNNs) with UWB signals for multiple road obstacle detection. Features are extracted from the time-frequency domain using the discrete wavelet transform and are forwarded to the Long short-term memory (LSTM) network. We evaluate our approach on the OLIMP dataset which includes various driving situations with complex environment and targets from several classes. The obtained results show that the LSTM-based system outperforms the other implemented related techniques in terms of obstacle detection.
Download

Paper Nr:	272
Title:	Market Impact in Trader-agents: Adding Multi-level Order-flow Imbalance-sensitivity to Automated Trading Systems
Authors:	Zhen Zhang and Dave Cliff
Abstract:	Financial markets populated by human traders often exhibit so-called “market impact”, where the prices quoted by traders move in the direction of anticipated change, before any transaction has taken place, as an immediate reaction to the arrival of a large (i.e., “block”) buy or sell order in the market: traders in the market know that a block buy order is likely to push the price up, and that a block sell order is likely to push the price down, and so they immediately adjust their quote-prices accordingly. In most major financial markets nowadays very many of the participants are “robot traders”, autonomous adaptive software agents, rather than humans. This paper addresses the question of how to give such trader-agents a reliable anticipatory sensitivity to block orders, such that markets populated entirely by robot traders also show market-impact effects. This is desirable because impact-sensitive trader-agents will get a better price for their transactions when block orders arrive, and because such traders can also be used for more accurate simulation models of real-world financial markets. In a 2019 publication Church & Cliff presented initial results from a simple deterministic robot trader, called ISHV, which was the first such trader-agent to exhibit this market impact effect. ISHV does this via monitoring a metric of imbalance between supply and demand in the market. The novel contributions of our paper are: (a) we critique the methods used by Church & Cliff, revealing them to be weak, and argue that a more robust measure of imbalance is required; (b) we argue for the use of multi-level order-flow imbalance (MLOFI: Xu et al., 2019) as a better basis for imbalance-sensitive robot trader-agents; and (c) we demonstrate the use of the more robust MLOFI measure in extending ISHV, and also the well-known AA and ZIP trading-agent algorithms (which have both been previously shown to consistently outperform human traders). Our results demonstrate that the new imbalance-sensitive trader-agents introduced in this paper do exhibit market impact effects, and hence are better-suited to operating in markets where impact is a factor of concern or interest, but do not suffer the weaknesses of the methods used by Church & Cliff. We have made the source-code for our work reported here freely available on GitHub.
Download

Paper Nr:	275
Title:	Multiple Pursuers TrailMax Algorithm for Dynamic Environments
Authors:	Azizkhon Afzalov, Ahmad Lotfi, Benjamin Inden and Mehmet E. Aydin
Abstract:	Multi-agent multi-target search problems, where the targets are capable of movement, require sophisticated algorithms for near-optimal performance. While there are several algorithms for agent control, comparatively less attention has been paid to near-optimal target behaviours. Here, a state-of-the-art algorithm for targets to avoid a single agent called TrailMax has been adapted to work within a multiple agents and multiple targets framework. The aim of the presented algorithm is to make the targets avoid capture as long as possible, if possible until timeout. Empirical analysis is performed on grid-based gaming benchmarks. The results suggest that Multiple Pursuers TrailMax reduces the agent success rate by up to 15% as compared to several previously used target control algorithms and increases the time until capture in successful runs.
Download

Paper Nr:	277
Title:	Measuring Inflation within Virtual Economies using Deep Reinforcement Learning
Authors:	Conor Stephens and Chris Exton
Abstract:	This paper proposes a framework for assessing economies within online multiplayer games without the need for extensive player testing and data collection. Players have identified numerous exploits in modern online games to further their collection of resources and items. A recent exploit within a game-economy would be in Animal Crossing New Horizons a multiplayer game released in 2020 which featured bugs that allowed users to generate infinite money (Sudario, 2020); this has impacted the player experience in multiple negative ways such as causing hyperinflation within the economy and scarcity of resources within the particular confines of any game. The framework proposed by this paper can aid game developers and designers when testing their game systems for potential exploits that could lead to issues within the larger game economies. Assessing game systems is possible by leveraging reinforcement learning agents to model player behaviour; this is shown and evaluated in a sample multiplayer game. This research is designed for game designers and developers to show how multi-agent reinforcement learning can help balance game economies. The project source code is open source and available at: https://github.com/Taikatou/economy research.
Download

Paper Nr:	279
Title:	Scalable Stochastic Path Planning under Congestion
Authors:	Kamilia Ahmadi and Vicki H. Allan
Abstract:	In this work, we propose a city scale path planning framework when edge weights are not fixed and are stochastically defined based on the mean and variance of travel time on each edge. Agents are car drivers who are moving from one point to another point in different time of the day/night. Agents can pursue two types of goals: first, the ones who are not willing to take risk and look for the path with highest probability of reaching destination before their desired arrival time, even if it may take them longer. The second group are the agents who are open to take a riskier decision if it helps them in having the shortest en-route time. In order to scale the path planning process and make it applicable to city scale, pre-computation and approximation has been used. The city graph is partitioned to smaller groups of nodes and each group is represented by one node which is called exemplar. For path planning queries, source and destination pair are connected to the respective exemplars correspond to the direction of source to destination and path between those exemplars is found. Paths are stored in distance oracles for different time slots of day/week in order to expedite the query time. Distance oracles are updated weekly in order to capture the recent changes in traffic. The results show that, this approach helps in having a scalable path finding framework which handles queries in real time while the approximate paths are at least 90 percent as good as the exact paths.
Download

Short Papers

Paper Nr:	9
Title:	Comparing Elementary Cellular Automata Classifications with a Convolutional Neural Network
Authors:	Thibaud Comelli, Frédéric Pinel and Pascal Bouvry
Abstract:	Elementary cellular automata (ECA) are simple dynamic systems which display complex behaviour from simple local interactions. The complex behaviour is apparent in the two-dimensional temporal evolution of a cellular automata, which can be viewed as an image composed of black and white pixels. The visual patterns within these images inspired several ECA classifications, aimed at matching the automatas’ properties to observed patterns, visual or statistical. In this paper, we quantitatively compare 11 ECA classifications. In contrast to the a priori logic behind a classification, we propose an a posteriori evaluation of a classification. The evaluation employs a convolutional neural network, trained to classify each ECA to its assigned class in a classification. The prediction accuracy indicates how well the convolutional neural network is able to learn the underlying classification logic, and reflects how well this classification logic clusters patterns in the temporal evolution. Results show different prediction accuracy (yet all above 85%), three classifications are very well captured by our simple convolutional neural network (accuracy above 99%), although trained on a small extract from the temporal evolution, and with little observations (100 per ECA, evolving 513 cells). In addition, we explain an unreported ”pathological” behaviour in two ECAs.
Download

Paper Nr:	17
Title:	The Power Index at Infinity: Weighted Voting in Sequential Infinite Anonymous Games
Authors:	Shereif Eid
Abstract:	After we describe the waiting queue problem, we identify a partially observable 2n+1-player voting game with only one pivotal player; the player at the n-1 order. Given the simplest rule of heterogeneity presented in this paper, we show that for any infinite sequential voting game of size 2n+1, a power index of size n is a good approximation of the power index at infinity, and it is difficult to achieve. Moreover, we show that the collective utility value of a coalition for a partially observable anonymous game given an equal distribution of weights is n²+n. This formula is developed for infinite sequential anonymous games using a stochastic process that yields a utility function in terms of the probability of the sequence and voting outcome of the coalition. Evidence from Wikidata editing sequences is presented and the results are compared for 10 coalitions.
Download

Paper Nr:	20
Title:	Inconsistency-tolerant Hierarchical Probabilistic Computation Tree Logic and Its Application to Model Checking
Authors:	Norihiro Kamide and Noriko Yamamoto
Abstract:	An inconsistency-tolerant hierarchical probabilistic computation tree logic (IHpCTL) is developed to establish a new extended model checking paradigm referred to as IHpCTL model checking, which is intended to verify randomized, open, large, and complex concurrent systems. The proposed IHpCTL is constructed based on several previously established extensions of the standard probabilistic temporal logic known as probabilistic computation tree logic (pCTL), which is widely used for probabilistic model checking. IHpCTL is shown to be embeddable into pCTL and is relatively decidable with respect to pCTL. This means that the decidability of pCTL with certain probability measures implies the decidability of IHpCTL. The results indicate that we can effectively reuse the previously proposed pCTL model-checking algorithms for IHpCTL model checking.
Download

Paper Nr:	21
Title:	Forecasting Air Pollution in Munich: A Comparison of MLR, ANFIS, and SVM
Authors:	Andreas Humpe, Lars Brehm and Holger Günzel
Abstract:	As motor vehicle air pollution is a serious health threat, there is a need for air quality forecasting to fulfil policy requirements, and lower traffic induced air pollution. This article compares the performance of multiple linear regressions, adaptive neuro-fuzzy inference systems, and support vector machines in predicting one-hour ahead particulate matter, nitrogen oxides and ozone concentration in the City of Munich between 2014 and 2018. The models are evaluated with different performance measures in-sample and out-of-sample. The results generally support earlier studies on forecasting air pollution and indicate that adaptive neuro-fuzzy inference systems have the highest predictive power in terms of R-square for all pollutants. Furthermore, ozone can be predicted best, whereas nitrogen oxides are the least predictive pollutants. One reason for the different predictability might be rooted in the short lifetime of nitrogen oxides compared to ozone. The results here should be of interest to regulators and municipal traffic managements alike who are interested in predicting air pollution and improve urban air quality.
Download

Paper Nr:	26
Title:	Design and Implementation of German Legal Decision Corpora
Authors:	Stefanie Urchs, Jelena Mitrović and Michael Granitzer
Abstract:	Law professionals are wordsmiths, their main tool is language. Therefore, the field of law produces a vast amount of written text. These texts have to be analysed, summarised, and used in the creation of new text, which is a task that reaches the limits of what is humanly possible. However, it is possible to automate this analysis by using Natural Language Processing techniques. To perform these techniques (annotated) text corpora are required. Unfortunately, publicly available (annotated) legal text corpora are rare. Even scarcer is the availability of (annotated) German legal text corpora. To meet this need for publicly available German legal text corpora this paper presents two German legal text corpora. The first corpus contains 32,748 decisions from 131 German courts, enriched with metadata. The second one is a subset of the first corpus and consists of 200 randomly chosen judgements. In these judgements a legal expert annotated the components conclusion, definition and subsumption of the German legal writing style Urteilsstil. Furthermore, the paper presents experiments on these corpora.
Download

Paper Nr:	35
Title:	Quantum Control for Error Correction using Mother Tee Optimization
Authors:	Wael Korani and Malek Mouhoub
Abstract:	Quantum Control Problem (QCP) for Error Correction (EC) is a significant issue that helps in producing an efficient quantum computer. The QCP for EC can be tackled using Stochastic Local Search (SLS) methods. However, these techniques might produce low quality results for large dimensional quantum systems. Lately, Nature-Inspired (NI) algorithms including different variants of Particle Swarm Optimization (PSO) and Deferential Evolution (DE) were implemented in several studies to tackle the QCP for EC, but the results were not promising. In this paper, we propose a quantum model that is built on our NI algorithm, called Mother Tree Optimization for QCP (MTO-QCP), to overcome the stagnation issue that the other methods suffer from. In order to assess the performance of MTO-QCP, we conducted several preliminary experiments to adjust our MTO parameters. In this regard, our MTO-QCP achieves high-fidelity (> 99.99%) for a Single-Shot (SS) three-qubit gate control at gate operation time of 26 ns. This recommended fidelity is an acceptable threshold fidelity for fault-tolerant Quantum Computing (QC) problems
Download

Paper Nr:	36
Title:	A Probabilistic Theory of Abductive Reasoning
Authors:	Nicolas A. Espinosa Dice, Megan L. Kaye, Hana Ahmed and George D. Montañez
Abstract:	We present an abductive search strategy that integrates creative abduction and probabilistic reasoning to produce plausible explanations for unexplained observations. Using a graphical model representation of abductive search, we introduce a heuristic approach to hypothesis generation, comparison, and selection. To identify creative and plausible explanations, we propose 1) applying novel structural similarity metrics to a search for simple explanations, and 2) optimizing for the probability of a hypothesis’ occurrence given known observations.
Download

Paper Nr:	37
Title:	Probabilistic (k,l)-Context-Sensitive Grammar Inference with Gibbs Sampling Applied to Chord Sequences
Authors:	Henrique B. Lopes and Alan Freitas
Abstract:	Grammatical inference in computer music provides us with valuable models for fields such as algorithmic composition, style modeling, and music theory analysis. Grammars with higher accuracy can lead to models that improve the performance of various tasks in these fields. Recent studies show that Hidden Markov Models can outperform Markov Models in terms of accuracy, but there are no significant differences between Hidden Markov Models and Probabilistic Context-Free Grammars (PCFGs). This paper applies a Gibbs Sampling algorithm to infer Probabilistic (k,l)-Context-Sensitive Grammars (P(k,l)CSGs) and presents an application of P(k,l)CSGs to model the generation of chord sequences. Our results show Gibbs Sampling and P(k,l)CSGs can improve on PCFGs and the Metropolis-Hastings algorithm with perplexity values that are 48% lower on average (p-value 0.0026).
Download

Paper Nr:	43
Title:	Manufacturing Control in Job Shop Environments with Reinforcement Learning
Authors:	Vladimir Samsonov, Marco Kemmerling, Maren Paegert, Daniel Lütticke, Frederick Sauermann, Andreas Gützlaff, Günther Schuh and Tobias Meisen
Abstract:	Computing solutions to job shop problems is a particularly challenging task due to the computational hardness of the underlying optimization problem as well as the often dynamic nature of given environments. To address such scheduling problems in a more flexible way, such that changing circumstances can be accommodated, we propose a reinforcement learning approach to solve job shop problems. As part of our approach, we propose a new reward shaping and devise a novel action space, from which a reinforcement learning agent can sample actions, which is independent of the job shop problem size. A number of experiments demonstrate that our approach outperforms commonly used scheduling heuristics with regard to the quality of the generated solutions. We further show that, once trained, the time required to compute solutions using our methodology increases less sharply as the problem size grows than exact solution methods making it especially suitable for online manufacturing control tasks.
Download

Paper Nr:	48
Title:	Optimal Distribution of CNN Computations on Edge and Cloud
Authors:	Paul A. Leroy and Toon Goedemé
Abstract:	In this paper we study the optimal distribution of CNN computations between an edge device and the cloud for a complex IoT application. We propose a pipeline in which we perform experiments with a Jetson Nano and a Raspberry Pi 3B+ as the edge device, and a T2.micro instance from Amazon EC2 as a cloud instance. To answer this generic question, we performed exhaustive experiments on a typical use case, a mobile camera- based street litter detection and mapping application based on a MobilenetV2 model. For our research, we split the computations of the CNN model and divided them over the edge and cloud instances using model partitioning, also including the edge-only and cloud-only configurations. We studied the influence of the specifications of the instances, the input size of the model, the partitioning location of the model and the available network bandwidth on the optimal split position. Depending on the choice of gaining either an economic or performance advantage, we can conclude that a balance between the choice of instances and the calculation mechanism used should be made.
Download

Paper Nr:	49
Title:	Why It is Hard to Find AI in SMEs: A Survey from the Practice and How to Promote It
Authors:	Andreas Bunte, Frank Richter and Rosanna Diovisalvi
Abstract:	AI seems to be an important aspect of Industry 4.0, which was introduced about 10 years ago. The main results of interviews about AI with 411 people from 68 companies have been summarized in this paper. Most of those companies were SMEs. Main challenges for the application of AI have been identified. Concrete solutions that can support the implementation and application of AI are presented. The need to adequately support AI in SMEs is underlined and specified.
Download

Paper Nr:	51
Title:	Markov Logic Network for Metaphor Set Expansion
Authors:	Jaya Pathak and Pratik Shah
Abstract:	Metaphor is a figure of speech, that allow us to understand a concept of a domain in terms of the other. One of the sub-problems related to the metaphor recognition is of metaphor set expansion. This in turn is an instance of information completion problem. We, in this work, propose an MLN based approach to address the problem of metaphor set expansion. The rules for metaphor set expansion are represented in the first order logic formulas. The rules are either soft or hard depending on the nature of the rules according to which corresponding logic formulas are then assigned weights. Many a times new metaphors are created based on usages of Is-A pair knowledge base. We, in this work model this phenomena by introducing appropriate predicates and formulas in clausal form. For experiments, we have used dataset from Microsoft concept graph consisting Is-A patterns. The experiments show that the weights for the formulas can be learnt using the training dataset. Moreover the formulas and their weights are easy to interpret and in-turn explains the inference results adequately. We believe that this is a first effort reported which uses MLN for metaphor set expansion.
Download

Paper Nr:	58
Title:	Using Game AI to Control a Simulated Economy
Authors:	Paul McCarlie and Aaron Hunter
Abstract:	We explore the use of Artificial Intelligence (AI) to manipulate a simulated economy. Towards this end, we present work in progress on a macroeconomic simulation that can be controlled by a game player. We view this simulation as a sort of serious game; it can be played as a competition, but it can also be an educational tool through which players learn both about economic principles and the behaviour of AI controllers. The main contribution of this paper is the comparative study of the effectiveness of different AI agents for the manipulation of a simulated environment. Focusing on AI approaches that are common in the gaming industry, we implement four players that use intelligent methods to control the simulation by trying to maximize the economic output. The aim of our work is to illustrate that simple methods from the game AI community can be used to control a complex economic simulation effectively. This work, therefore, supports the common position in the gaming community that simple character-based AI methods can produce competitive game play even for complex tasks. Moreover, we demonstrate that, in the case of this particular simulation, a rule-based reasoner outperforms more sophisticated AI agents.
Download

Paper Nr:	62
Title:	A Quality Framework for Automated Planning Knowledge Models
Authors:	Mauro Vallati and Thomas L. Mccluskey
Abstract:	Automated planning is a prominent Artificial Intelligence challenge, as well as a requirement for intelligent autonomous agents. A crucial aspect of automated planning is the knowledge model, that includes the relevant aspects of the application domain and of a problem instance to be solved. Despite the fact that the quality of the model has a strong influence on the resulting planning application, the notion of quality for automated planning knowledge models is not well understood, and the engineering process in building such models is still mainly an ad-hoc process. In order to develop systematic processes that support a more comprehensive notion of quality, this paper, building on existing frameworks proposed for general conceptual models, introduces a quality framework specifically focused on automated planning knowledge models.
Download

Paper Nr:	69
Title:	Smartphone Glass Inspection System
Authors:	Sergey Turko, Liudmila Burmak, Ilya Malyshev, Stanislav Shtykov, Mikhail Popov, Pavel Filimonov, Alexandr Aspidov and Andrei Shcherbinin
Abstract:	In this paper we address the problem of detection and discrimination of defects on smartphone cover glass. Specifically, scratches and scratch-like defects. An automatic detection system which allows to detect scratches on the whole surface of a smartphone’s cover glass without human participation is developed. The glass sample is illuminated sequentially from several directions using a special ring illumination system and a camera takes a dark-field image at each illumination state. The captured images show a variation of the defect image intensity depending on the illumination direction. We present a pipeline of detecting scratches on images obtained by our system using convolutional neural networks (CNN) and particularly U-net-like architecture. We considered the scratch detection problem as a semantic segmentation task. The novel loss technique for solving the problem of imbalance, sparsity and low representability of data is presented. The proposed technique solves two tasks simultaneously: segmentation and reconstruction of the provided image. Also, we suggest a nested convolution kernels to overcome the problem of overfitting and to extend the receptive field of the CNN without increasing trainable weights.
Download

Paper Nr:	78
Title:	Using Deep Learning for Trajectory Classification
Authors:	Nicksson C. A. de Freitas, Ticiana L. Coelho da Silva, José A. Fernandes de Macêdo, Leopoldo Melo Junior and Matheus G. Cordeiro
Abstract:	The ubiquity of GPS-enabled smartphones and automotive navigation systems connected to the Internet allows us to monitor, collect, and analyze large trajectory data streams in real-time. Trajectory classification is an efficient way to analyze trajectory, consisting of building a prediction model to classify a new trajectory (or sub-trajectory) in a single-class or multi-class. The classification trajectory problem is challenging because of the massive volume of trajectory data, the complexity associated with the data representation, the sparse nature of the spatio-temporal points, the multidimensionality, and the number of classes can be much larger than the number of motion patterns. Machine learning methods can handle trajectories, but they demand a feature extraction process, and they suffer from the curse of dimensionality. On the other hand, more recent Deep Learning models emerged to link trajectories to their generating users. Although they minimize the sparsity problem by representing the input data as an embedding vector, these models limit themselves to deal with multidimensional data. In this paper, we propose DeepeST (Deep Learning for Sub-Trajectory classification) to identify the category from a large number of sub-trajectories extracted from GPS services and check-ins data. DeepeST employs a Recurrent Neural Network (RNN), both LSTM and Bi-directional LSTM (BLSTM), which operates on the low-dimensional to learn the underlying category. We tackled the classification problem and conducted experiments on three real datasets with trajectories from GPS services and check-ins. We show that DeepeST outperforms machine learning approaches and deep learning approaches from state-of-the-art.
Download

Paper Nr:	79
Title:	In-car Damage Dirt and Stain Estimation with RGB Images
Authors:	Sandra Dixe, João Leite, Sahar Azadi, Pedro Faria, José Mendes, Jaime C. Fonseca and João Borges
Abstract:	Shared autonomous vehicles (SAV) numbers are going to increase over the next years. The absence of human driver will create a new paradigm for in-car safety. This paper addresses the problem, presenting a monitoring system capable of estimating the state of the car interior, namely the presence of damage, dirt and stains. We propose the use of Semantic Segmentation methods to perform appropriate pixel-wise classification of certain textures found in the car’s cabin as defect classes. Two methods, U-Net and DeepLabV3+, were trained and tested for different hiper-parameter and ablation scenarios, using RGB images. To be able to test and validate these approaches an In-car dataset was created, comprised by 1861 samples from 78 cars, and than splitted in 1303 train, 186 validation and 372 test RGB images. DeepLabV3+ showed promissing results, achieving an average accuracy for good, damage, stain and dirt of 77.17%, 58.60%, 65.81% and 68.82%, respectively.
Download

Paper Nr:	80
Title:	A Crowdsourcing Methodology for Improved Geographic Focus Identification of News-stories
Authors:	Christos Rodosthenous and Loizos Michael
Abstract:	Past work on the task of identifying the geographic focus of news-stories has established that state-of-the-art performance can be achieved by using existing crowdsourced knowledge-bases. In this work we demonstrate that a further refinement of those knowledge-bases through an additional round of crowdsourcing can lead to improved performance on the aforementioned task. Our proposed methodology views existing knowledge-bases as collections of arguments in support of particular inferences in terms of the geographic focus of a given news-story. The refinement that we propose is to associate these arguments with weights — computed through crowdsourcing — in terms of how strongly they support their inference. The empirical results that we present establish the superior performance of this approach compared to the one using the original knowledge-base.
Download

Paper Nr:	82
Title:	Generating Reactive Robots’ Behaviors using Genetic Algorithms
Authors:	Jesus Savage, Stalin Muñoz, Luis Contreras, Mauricio Matamoros, Marco Negrete, Carlos Rivera, Gerald Steinbabuer, Oscar Fuentes and Hiroyuki Okada
Abstract:	In this paper, we analize and benchmark three genetically-evolved reactive obstacle-avoidance behaviors for mobile robots. We buit these behaviors with an optimization process using genetic algorithms to find the one allowing a mobile robot to best reactively avoid obstacles while moving towards its destination. We compare three approaches, the first one is a standard method based on potential fields, the second one uses on finite state machines (FSM), and the last one relies on HMM-based probabilistic finite state machines (PFSM). We trained the behaviors in simulated environments to obtain the optimizated behaviors and compared them to show that the evolved FSM approach outperforms the other two techniques.
Download

Paper Nr:	84
Title:	An Improved Cuckoo Search Algorithm for Multiple Odor Sources Localization
Authors:	Yuqing Wu and Zhipu Wang
Abstract:	This work presents an improved Cuckoo Search Algorithm (CSA) for multiple odor sources localization. The idea of forbidden areas is introduced to the CSA as territories of the cuckoo colonies, preventing the cuckoos from being trapped into local optimal solutions. A source is declared when a certain number of cuckoos are located in close proximity with each other, and a territory is formed around the declared source centered at the local best among those cuckoos. When territories overlap, they are merged into one territory to prevent the same source from being found multiple times. Simulation results show that the proposed method can locate multiple odor sources with high accuracy.
Download

Paper Nr:	90
Title:	Modeling and Simulation of Associative Reasoning
Authors:	Jiří Jelínek
Abstract:	Modeling human behavior is a popular area of research. Special attention is then focused on activities related to knowledge processing. It is the knowledge that has a fundamental influence on an individual's decision-making and its dynamics. The subject of research is both the representation of knowledge and the procedures of their processing. The processing also comprises associative reasoning. Associations significantly influence the knowledge base used in processing stimuli and thus participate in creating a knowledge context that is further used for knowledge derivation and decision making. This paper focuses on the area of associative knowledge processing. There are already classical approaches associated with developing probabilistic neural networks, which can also be used with modifications at a higher abstraction level. This paper aims to show that associative processing of knowledge can be described with these approaches and simulated. The article will present a possible implementation of the model of knowledge storage and associative processing on the individual's knowledge base. The behavior of this model will be demonstrated in experiments.
Download

Paper Nr:	92
Title:	Binary Classification: Counterbalancing Class Imbalance by Applying Regression Models in Combination with One-sided Label Shifts
Authors:	Peter Bellmann, Heinke Hihn, Daniel A. Braun and Friedhelm Schwenker
Abstract:	In many real-world pattern recognition scenarios, such as in medical applications, the corresponding classification tasks can be of an imbalanced nature. In the current study, we focus on binary, imbalanced classification tasks, i.e. binary classification tasks in which one of the two classes is under-represented (minority class) in comparison to the other class (majority class). In the literature, many different approaches have been proposed, such as under- or oversampling, to counter class imbalance. In the current work, we introduce a novel method, which addresses the issues of class imbalance. To this end, we first transfer the binary classification task to an equivalent regression task. Subsequently, we generate a set of negative and positive target labels, such that the corresponding regression task becomes balanced, with respect to the redefined target label set. We evaluate our approach on a number of publicly available data sets in combination with Support Vector Machines. Moreover, we compare our proposed method to one of the most popular oversampling techniques (SMOTE). Based on the detailed discussion of the presented outcomes of our experimental evaluation, we provide promising ideas for future research directions.
Download

Paper Nr:	95
Title:	Real-time Active Vision for a Humanoid Soccer Robot using Deep Reinforcement Learning
Authors:	Soheil Khatibi, Meisam Teimouri and Mahdi Rezaei
Abstract:	In this paper, we present an active vision method using a deep reinforcement learning approach for a humanoid soccer-playing robot. The proposed method adaptively optimises the viewpoint of the robot to acquire the most useful landmarks for self-localisation while keeping the ball into its viewpoint. Active vision is critical for humanoid decision-maker robots with a limited field of view. To deal with active vision problem, several probabilistic entropy-based approaches have previously been proposed which are highly dependent on the accuracy of the self-localisation model. However, in this research, we formulate the problem as an episodic reinforcement learning problem and employ a Deep Q-learning method to solve it. The proposed network only requires the raw images of the camera to move the robot’s head toward the best viewpoint. The model shows a very competitive rate of 80% success rate in achieving the best viewpoint. We implemented the proposed method on a humanoid robot simulated in Webots simulator. Our evaluations and experimental show that the proposed method outperforms the entropy-based methods in the RoboCup context, in cases with high self-localisation errors.
Download

Paper Nr:	100
Title:	Post-hoc Explanation using a Mimic Rule for Numerical Data
Authors:	Kohei Asano and Jinhee Chun
Abstract:	We propose a novel rule-based explanation method for an arbitrary pre-trained machine learning model. Generally, machine learning models make black-box decisions that are not easy to explain the logical reasons to derive them. Therefore, it is important to develop a tool that gives reasons for the model’s decision. Some studies have tackled the solution of this problem by approximating an explained model with an interpretable model. Although these methods provide logical reasons for a model’s decision, a wrong explanation sometimes occurs. To resolve the issue, we define a rule model for the explanation, called a mimic rule, which behaves similarly in the model in its region. We obtain a mimic rule that can explain the large area of the numerical input space by maximizing the region. Through experimentation, we compare our method to earlier methods. Then we show that our method often improves local fidelity.
Download

Paper Nr:	104
Title:	Traffic Sign Recognition System based on Belief Functions Theory
Authors:	Nesrine Triki, Mohamed Ksantini and Mohamed Karray
Abstract:	Advanced Driver Assistance Systems (ADAS) have a strong interest in road safety. This type of assistance can be very useful for collision warning systems, blind spot detection and track maintenance assistance. Traffic Sign Recognition system is one of the most important ADAS technologies based on artificial intelligence methodologies where we obtain efficient solutions that can alert and assist the driver and, in specific cases, accelerate, slow down or stop the vehicle. In this work, we will improve the effectiveness and the efficiency of machine learning classifiers on traffic signs recognition process in order to satisfy ADAS reliability and safety standards. Hence, we will use MLP, SVM, Random Forest (RF) and KNN classifiers on our traffic sign dataset first, each classifier apart then, by fusing them using the Dempster-Shafer (DS) theory of belief functions. Experimental results confirm that by combining machine learning classifiers we obtain a significant improvement of accuracy rate compared to using classifiers independently.
Download

Paper Nr:	111
Title:	Contract Metadata Identification in Czech Scanned Documents
Authors:	Hien T. Ha, Aleš Horák and Minh T. Bui
Abstract:	Although nowadays digital-born documents are generally prevalent, exchange of business documents often consists in processing their scanned image form as a general human-readable format with one-to-one correspondence to paper documents. Bulk processing of such scanned documents then requires human intervention to extract and enter the main document metadata. In this paper, we present the design and evaluation of a contract processing module in the OCRMiner system. The information extraction process allows to combine layout properties with text analysis as input to a rule-based extraction with confidence score propagation. The first results are evaluated with public Czech contract documents reaching the item extraction accuracy of almost 88%.
Download

Paper Nr:	114
Title:	Improvements to Increase the Efficiency of the AlphaZero Algorithm: A Case Study in the Game ’Connect 4’
Authors:	Colin Clausen, Simon Reichhuber, Ingo Thomsen and Sven Tomforde
Abstract:	AlphaZero is a recent approach to self-teaching gameplay without the need for human expertise. It suffers from the massive computation and hardware requirements, which are responsible for the reduced applicability of the approach. This paper focuses on possible improvements with the goal to reduce the required computation resources. We propose and investigate three modifications: We model the self-learning phase as an evolutionary process, study the game process as a tree and use network-internal features as auxiliary targets. Then behaviour and performance of these modifications are evaluated in the game Connect 4 as a test scenario.
Download

Paper Nr:	115
Title:	Sentence Boundary Detection in German Legal Documents
Authors:	Ingo Glaser, Sebastian Moser and Florian Matthes
Abstract:	Sentence boundary detection on German legal texts is a task which standardized NLP-systems have little or no ability to handle, since they are sometimes overburdened by more complex structures such as lists, paragraph structures and citations. In this paper we evaluate the performance of these systems and adapt methods directly to the legal domain. We created an annotated dataset with over 50,000 sentences consisting of various German legal documents which can be utilized for further research within the community. Our neural networks and conditional random fields models show significantly higher performances on this data than the tested, already existing systems. Thus this paper contradicts the assumption that the problem of segmenting sentences is already solved.
Download

Paper Nr:	120
Title:	Recipe Enrichment: Knowledge Required for a Cooking Assistant
Authors:	Nils Neumann and Sven Wachsmuth
Abstract:	The preparation of a meal consisting of multiple components with different timing and critical synchronization points is a complex task. An automated system assisting a human in the preparation process needs to track the progress state, prompt the next recipe steps, control kitchen devices, monitor the final preparation time, and deal with process deviations. This requires a detailed process representation including knowledge about states with critical timing, control signals of devices, preparation steps and cooking times of ingredients, and necessary user attention. At the same time, the system should be flexible enough to allow the free combination of component recipes and kitchen devices to support the preparation of complete menus independent of a specific kitchen setup. To meet these requirements, we propose a method to automatically enrich simple component recipes with process-specific information. The resulting detailed process description can be processed by standard scheduling algorithms to sequence the preparation steps of complex meals. The control information for kitchen devices is already included in the process description, so that monitoring of the progress becomes possible. The reasoning process is driven by so-called action templates that allow to decouple knowledge on recipes, ingredients, and kitchen devices in seperate re-usable knowledge bases.
Download

Paper Nr:	122
Title:	Sensorless Coil Temperature Measurements using Neural Networks for Voltage Control
Authors:	Linus Taenzer, Chafic Abu-Antoun and Jasmin Smajic
Abstract:	Voltage and current measurement data based deep learning as a method to conduct sensorless coil temperature prediction of an embedded linear induction actuator is being proposed and validated in this work. Generated numerical data from Finite Element field simulations are used to train a neural network which in turn predicts temperatures at non-accessible places e.g. at an embedded coil. The network is demonstrated and the comparison to experimental data shows the potential of virtual sensing. Even though the number of physical sensors have increased enormously in the last decades, the measurement of desired temperatures at certain locations is limited by accessibility and by the application itself, for example, if a coil is used as a moving part in an actuator. This work proposes an indirect method based on measurable quantities in the device, such as voltage and current, to quantify precisely temperatures and hot spots in sensitive parts of the device. As high temperatures can have a huge effect on the device’s performance, a controllable voltage to compensate the performance reduction instantaneously is desired. Applications based on the principle of an inductive linear actuator show a strong performance dependency on the temperature of the conducting material or coil. The authors present an Artificially Intelligent voltage controller to achieve the desired performance based on measurable variables in the device and supported by sensorless methods like temperature prediction with Artificial Intelligence (AI).
Download

Paper Nr:	123
Title:	A Two Step Fine-tuning Approach for Text Recognition on Identity Documents
Authors:	Francesco Visalli, Antonio Patrizio and Massimo Ruffolo
Abstract:	Manually extracting data from documents for digitization is a long, tedious and error-prone job. In recent years, technologies capable of automating these processes are gaining ground and managing to obtain surprising results. Research in this field is driven by the strong interest of organizations that have identified how the automation of data entry leads to a reduction in working time and a speed-up of business processes. Documents of interest are heterogeneous in format and content. These can be natively machine readable or not when they are images obtained by scanning paper. Documents in image format require pre-processing before applying information extraction. A typical pre-processing pipeline consists of two steps: text detection and text recognition. This work proposes a two step fine-tuning approach for text recognition in Italian identity documents based on Scene Text Recognition networks. Experiments show promising results.
Download

Paper Nr:	128
Title:	Receiving Messages in Their Correct Order: Analyzing Broadcast Protocols in Dynamic Epistemic Logics
Authors:	Spandan Das and Sujata Ghosh
Abstract:	In this paper we analyse distributed protocols in a logical framework. In particular, we provide a dynamic epistemic logic analysis of certain broadcast protocols, viz. Birman-Schiper-Stephenson protocol and Lamport’s mutual exclusion protocol. In the process, we provide correctness proofs of these protocols via knowledge-based reasoning. We also provide a detailed algorithmic analysis of the logical modeling of these protocols.
Download

Paper Nr:	130
Title:	Novelty Detection in Physical Activity
Authors:	Bernardo Leite, Amr Abdalrahman, João Castro, Julieta Frade, João Moreira and Carlos Soares
Abstract:	Artificial Intelligence (AI) is continuously improving several aspects of our daily lives. There has been a great use of gadgets & monitoring devices for health and physical activity monitoring. Thus, by analyzing large amounts of data and applying Machine Learning (ML) techniques, we have been able to infer fruitful conclusions in various contexts. Activity Recognition is one of them, in which it is possible to recognize and monitor our daily actions. The main focus of the traditional systems is only to detect pre-established activities according to the previously configured parameters, and not to detect novel ones. However, when applying activity recognizers in real-world applications, it is necessary to detect new activities that were not considered during the training of the model. We propose a method for Novelty Detection in the context of physical activity. Our solution is based on the establishment of a threshold confidence value, which determines whether an activity is novel or not. We built and train our models by experimenting with three different algorithms and four threshold values. The best results were obtained by using the Random Forest algorithm with a threshold value of 0.8, resulting in 90.9% of accuracy and 85.1% for precision.
Download

Paper Nr:	132
Title:	Explainable Reinforcement Learning for Longitudinal Control
Authors:	Roman Liessner, Jan Dohmen and Marco Wiering
Abstract:	Deep Reinforcement Learning (DRL) has the potential to surpass the existing state of the art in various practical applications. However, as long as learned strategies and performed decisions are difficult to interpret, DRL will not find its way into safety-relevant fields of application. SHAP values are an approach to overcome this problem. It is expected that the addition of these values to DRL provides an improved understanding of the learned action-selection policy. In this paper, the application of a SHAP method for DRL is demonstrated by means of the OpenAI Gym LongiControl Environment. In this problem, the agent drives an autonomous vehicle under consideration of speed limits in a single lane route. The controls learned with a DDPG algorithm are interpreted by a novel approach combining learned actions and SHAP values. The proposed RL-SHAP representation makes it possible to observe in every time step which features have a positive or negative effect on the selected action and which influences are negligible. The results show that RL-SHAP values are a suitable approach to interpret the decisions of the agent.
Download

Paper Nr:	136
Title:	Novel Radar-based Gesture Recognition System using Optimized CNN-LSTM Deep Neural Network for Low-power Microcomputer Platform
Authors:	Mateusz Chmurski and Mariusz Zubert
Abstract:	The goal of the embedded hand gesture recognition based on a radar sensor is to improve a human-machine interface, while taking into consideration privacy issues of camera sensors. In addition, the system has to be deployable on a low-power microcomputer for the applicability in broadly defined IoT and smart home solutions. Currently available gesture sensing solutions are ineffective in terms of low-power consumption what prevents them from the deployment on the low-power microcomputers. Recent advances exhibit a potential of deep learning models for a gesture classification whereas they are still limited to high-performance hardware. Embedded microcomputers are constrained in terms of memory, CPU clock speed and GPU performance. These limitations imply a topology design problem that is addressed in this work. Moreover, this research project proposes an alternative signal processing approach – using the continuous wavelet transform, which enables us to see the distribution of frequencies formed by every gesture. The newly proposed neural network topology performs equally well compared to the state of the art neural networks, however it needs only 54.6% of memory and it needs 20% of time to perform inference relative to the state of the art models. This dedicated neural network architecture allows for the deployment on resource constrained microcomputers, thus enabling a human-machine interface implementation on the embedded devices. Our system achieved an overall accuracy of 95.05% on earlier unseen data.
Download

Paper Nr:	138
Title:	Dancing Guide: Near Realtime Audio Classification of Live Dance Music on Smartphones
Authors:	Alexander K. Seewald
Abstract:	Between 2008 and 2014 we developed and deployed a live music classification system, Dancing Guide, to be run on Android and iPhones mobile phones in near realtime. Although internet access was needed to send feedback and classifications to the server for statistical purposes, the music classification system also worked offline without any loss in accuracy or speed. This is essential since in most discos and dancing schools, internet access is spotty at best. During the seven years of the project, the app was available both for iPhone and Android, initially in German and English, but later – thanks to volunteer translations – also in Czech, Spanish, French, Italian, Japanese, Korean, Dutch, Polish, Portuguese, Russian and Traditional Chinese. Measured by user feedback, we achieved an accuracy of roughly 73% at a coverage of 61%. While the accuracy is comparable to initial estimates using cross-validation, the coverage is much worse. Background noise – which we did not model – or the limited feature set may have been responsible. We retrained the system several times, however performance did not always improve, so we sometimes left the previously trained system in place. In the end, the limited feature set which was initially chosen prevented further improvement of coverage and accuracy, and we stopped further development.
Download

Paper Nr:	141
Title:	A QUBO Model to the Tail Assignment Problem
Authors:	Luis N. Martins, Ana P. Rocha and Antonio M. Castro
Abstract:	Tail Assignment is the problem of allocating individual aircraft to a set of flights subject to multiple constraints while optimising an objective function, such as operational costs. Given the enormous amount of possibilities and constraints involved, this problem has been a case study over the last decade. Many solutions have emerged using classical computing, but with limitations. Quantum Annealing (QA) is a heuristic technique to solve combinatorial optimisation problems by finding global minimum energy levels over an energy landscape using quantum mechanics. In this study, Tail Assignment Problem was framed as a Quadratic Unconstrained Binary Optimisation (QUBO) model and was solved using a classical and two hybrid solvers. The considered hybrid solvers made use of the D-Wave 2000Q quantum annealer. Tests were run based on extractions from real-world data, analysing, empirically, the performance of the implementation in terms of quality (i.e., the lowest operational costs) of the obtained solutions. We concluded that, for the considered datasets, there was a higher probability of obtaining better solutions for this problem using one of the hybrid solvers when compared with a classical heuristic algorithm such as Simulated Annealing (SA).
Download

Paper Nr:	144
Title:	On-demand Robotic Fleet Routing in Capacitated Networks with Time-varying Transportation Demand
Authors:	Martin Schaefer, Michal Čáp, David Fiedler and Jiří Vokřínek
Abstract:	In large-scale automated mobility-on-demand systems, the fleet manager is able to assign routes to individual automated vehicles in a way that minimizes formation of congestion. We formalize the problem of on-demand fleet routing in capacitated networks with time-varying demand. We demonstrate the limits of application of the steady-state flows approach in systems with time-varying demand and formulate a linear program to compute congestion-free routes for the vehicles in capacitated networks under time-varying demand. We evaluate the proposed approach in the simulation of a simplified, but characteristic illustrative example. The experiment reveals that the proposed routing approach can route 42% more traffic in congestion-free regime than the steady-state flow approach through the same network.
Download

Paper Nr:	158
Title:	Congestion-Aware Stochastic Path Planning and Its Applications in Real World Navigation
Authors:	Kamilia Ahmadi and Vicki H. Allan
Abstract:	In the realm of path planning, algorithms use edge weights in order to select the best path from an origin point to a specific target. This research focuses on the case where the edge weights are not fixed. Depending on the time of day/week, edge weights may change due to the congestion through the network. The best path is the path with minimum expected cost. The interpretation of best path depends on the point of view of car drivers. We model two different goals: 1) drivers who look for the path with the highest probability of reaching the destination before the deadline and 2) the drivers who look for the best time slot to leave in order to have a smallest travel time while they meet the deadline. Both of the goals are modelled based on the cost of the path which is highly dependent on the level of congestion in the network. Minimizing the paths’ cost helps in reducing traffic in the city, alleviates air pollution, and reduces fuel consumption. Findings show that using our proposed intelligent path planning algorithm which satisfies users’ goals and picks the least congested path is more cost efficient than picking the shortest-length path. Also, we show how agents’ goals and selection of cost function impacts paths’ choice.
Download

Paper Nr:	173
Title:	Non-alternative Artificial and Natural Intelligences Partnership for Put up Shoot in and Return to Rational Co-evolution with Noosphere
Authors:	Nicolay Vasilyev, Vladimir Gromyko and Stanislav Anosov
Abstract:	Sophisticated informational resources are poorly supported by semantic assistance. To untwist intellectual processes under inter-disciplinary activity in computer systems, life-long partnership with deep-learned artificial intelligence (DL IA) is needed. Resolving universalities problem to acquire knowledge self-obviousness helps a person to enter hermeneutic circle of noosphere by means of rational auto-poiesis. Otherwise, genus achievements will remain beyond the powers of the person. Purposeful tutoring allows transcendental apperception of the third world on the base of super sensual mathematical meanings. System axiomatic method (AMS) secures true profiling of computer systems and DL IA building. Its adaptive assistance to subject’s self-reflection in intuitive – discursive forms will apply anthropogenesis laws to encourage one’s rational consciousness self – building. DL IA implantation leans on cogno-ontological knowledge base mining. The latter is to be expressed in language of categories playing the role of mathesis universalis to embrace the variety of theories in their comparative description. Semantic glottogenesis keeps up the outlook of cognogenesis and will maintain communication on ideas level. The technology is based on reduction method applied to the person as functional system existing in anthropogenic nature.
Download

Paper Nr:	177
Title:	ML-based Decision Support for CSP Modelling with Regular Membership and Table Constraints
Authors:	Sven Löffler, Ilja Becker and Petra Hofstedt
Abstract:	The regular membership and the table constraints are very powerful constraints which allow it to substitute every other constraint in a constraint satisfaction problem. Both constraints can be used very flexible in a huge amount of problems. The main question we want to answer with this paper is, when is it faster to use the regular membership constraint, and when the table constraint. We use a machine learning approach for such a prediction based on propagation times. As learning input it takes randomly generated constraint problems, each containing exactly one table resp. one regular membership constraint. The evaluation of the resulting decision tool with specific but randomly generated CSPs shows the usefulness of our approach.
Download

Paper Nr:	178
Title:	A Best-first Backward-chaining Search Strategy based on Learned Predicate Representations
Authors:	Alexander Sakharov
Abstract:	Inference methods for first-order logic are widely used in knowledge base engines. These methods are powerful but slow in general. Neural networks make it possible to rapidly approximate the truth values of ground atoms. A hybrid neural-symbolic inference method is proposed in this paper. It is a best-first search strategy for backward chaining. The strategy is based on neural approximations of the truth values of literals. This method is precise and the results are explainable. It speeds up inference by reducing backtracking.
Download

Paper Nr:	179
Title:	Autoencoder Watchdog Outlier Detection for Classifiers
Authors:	Justin Bui and Robert J. Marks II
Abstract:	Neural networks have often been described as black boxes. A generic neural network trained to differentiate between kittens and puppies will classify a picture of a kumquat as a kitten or a puppy. An autoencoder watchdog screens trained classifier/regression machine input candidates before processing, e.g. to first test whether the neural network input is a puppy or a kitten. Preliminary results are presented using convolutional neural networks and convolutional autoencoder watchdogs using MNIST images.
Download

Paper Nr:	180
Title:	Adaptive Semantic Routing in Dynamic Environments
Authors:	Stefan Jakob, Harun Baraki, Alexander Jahl, Eric D. Nyakam Chiadjeu, Yasin Alhamwy and Kurt Geihs
Abstract:	Digital cities increasingly depend on IT and communication infrastructure. They can be seen as large distributed systems consisting of heterogeneous and autonomous participants. Especially during emergency situations and natural disasters, crucial knowledge about injured people and damaged infrastructure has to be discoverable by rescuers. Since knowledge discovery is about finding semantic information, IP-based routing mechanisms have to rely on flooding as they have to query each node to locate matching information. This leads to additional stress on the potentially damaged communication infrastructure. Therefore, we propose a semantic routing mechanism tailored for loosely-coupled networks that can dynamically change and are unstructured. By providing a multi-agent system, devices are enabled as part of the network to support the communication infrastructure. The focus of the network is set on discovering semantically structured knowledge, which can change at run-time. The main contributions of this paper are routing tables that incorporate semantic information to aggregate semantically close knowledge. They are only updated if new knowledge is available or new aggregates are created, to avoid network flooding. Based on a search & rescue scenario, we explain our new routing and update algorithms. Finally, we discuss the message complexity of the routing mechanism and the suitability of the used knowledge representation.
Download

Paper Nr:	182
Title:	Temporal Transfer Learning for Ozone Prediction based on CNN-LSTM Model
Authors:	Tuo Deng, Astrid Manders, Arjo Segers, Yanqin Bai and Hai X. Lin
Abstract:	Tropospheric ozone is a secondary pollutant which can affect human health and plant growth. In this paper, we investigated transferred convolutional neural network long short-term memory (TL-CNN-LSTM) model to predict ozone concentration. Hourly CNN-LSTM model is used to extract features and predict ozone for next hour, which is superior to commonly used models in previous studies. In the daily ozone prediction model, prediction over a large time-scale requires more data, however, only limited data are available, which causes the CNN-LSTM model to fail to accurately predict. Network-based transfer learning methods based on hourly models can obtain information from smaller temporal resolution. It can reduce prediction errors and shorten run time for model training. However, for extreme cases where the amount of data is severely insufficient, transfer learning based on smaller time scale cannot improve model prediction accuracy.
Download

Paper Nr:	185
Title:	Introducing the Hidden Neural Markov Chain Framework
Authors:	Elie Azeraf, Emmanuel Monfrini, Emmanuel Vignon and Wojciech Pieczynski
Abstract:	Nowadays, neural network models achieve state-of-the-art results in many areas as computer vision or speech processing. For sequential data, especially for Natural Language Processing (NLP) tasks, Recurrent Neural Networks (RNNs) and their extensions, the Long Short Term Memory (LSTM) network and the Gated Recurrent Unit (GRU), are among the most used models, having a “term-to-term” sequence processing. However, if many works create extensions and improvements of the RNN, few have focused on developing other ways for sequential data processing with neural networks in a “term-to-term” way. This paper proposes the original Hidden Neural Markov Chain (HNMC) framework, a new family of sequential neural models. They are not based on the RNN but on the Hidden Markov Model (HMM), a probabilistic graphical model. This neural extension is possible thanks to the recent Entropic Forward-Backward algorithm for HMM restoration. We propose three different models: the classic HNMC, the HNMC2, and the HNMC-CN. After describing our models’ whole construction, we compare them with classic RNN and Bidirectional RNN (BiRNN) models for some sequence labeling tasks: Chunking, Part-Of-Speech Tagging, and Named Entity Recognition. For every experiment, whatever the architecture or the embedding method used, one of our proposed models has the best results. It shows this new neural sequential framework’s potential, which can open the way to new models, and might eventually compete with the prevalent BiLSTM and BiGRU.
Download

Paper Nr:	187
Title:	On the Relevance of Extracting Macro-operators with Non-adjacent Actions: Does It Matter?
Authors:	Sandra Castellanos-Paez, Romain Rombourg and Philippe Lalanda
Abstract:	Understanding the role that plays the extraction phase on identifying potential macro candidates to augment a domain is critical. In this paper, we present a method to analyse the link between extracting macro-operators from non-adjacent actions and the correctness of (1) the frequency and (2) the number of occurrences per plan. We carried out experiments using our method on five benchmark domains and three different planners. We found that extracting macro-operators with only adjacent actions leads to important errors in macro-operator frequency and occurrences per plan.
Download

Paper Nr:	189
Title:	Emotion Recognition through Voting on Expressions in Multiple Facial Regions
Authors:	Ekanshi Agrawal, Jabez Christopher and Vasan Arunachalam
Abstract:	Facial Expressions are a key part of human behavior, and a way to express oneself and communicate with others. Multiple groups of muscles, belonging to different parts of the face, work together to form an expression. It is quite possible that the emotions being expressed by the region around the eyes and that around the mouth, don’t seem to agree with each other, but may agree with the overall expression when the entire face is considered. In such a case, it would be inconsiderate to focus on a particular region of the face only. This study evaluates expressions in three regions of the face (eyes, mouth, and the entire face) and records the expression reported by the majority. The data consists of images labelled with intensities of Action Units in three regions – eyes, mouth, and the entire face – for eight expressions. Six classifiers are used to determine the expression in the images. Each classifier is trained on all three regions separately, and then tested to determine an emotion label separately for each of the three regions of a test image. The image is finally labelled with the emotion present in at least two (or majority) of the three regions. Average performance over five stratified train-test splits it taken. In this regard, the Gradient Boost Classifier performs the best with an average accuracy of 94%, followed closely by Random Forest Classifier at 92%. The results and findings of this study will prove helpful in current situations where faces are partially visible and/or certain parts of the face are not captured clearly.
Download

Paper Nr:	198
Title:	A Comparison of Bayesian and Frequentist Approaches for the Case of Accident and Safety Analysis, as a Precept for All AI Expert Models
Authors:	Moldir Zholdasbayeva and Vasilios Zarikas
Abstract:	Statistical modelling techniques are widely used in accident studies. It is a well-known fact that frequentist statistical approach includes hypothesis testing, correlations, and probabilistic inferences. Bayesian networks, which belong to the set of advanced AI techniques, perform advanced calculations related to diagnostics, prediction and causal inference. The aim of the current work is to present a comparison of Bayesian and Regression approaches for safety analysis. For this, both advantages and disadvantages of two modelling approaches were studied. The results indicated that the precision of Bayesian network was higher than that of the ordinal regression model. However, regression analysis can also provide understanding of the information hidden in data. The two approaches may suggest different significant explanatory factors/causes, and this always should be taken into consideration. The obtained outcomes from this analysis will contribute to the existing literature on safety science and accident analysis.
Download

Paper Nr:	199
Title:	A Hybrid Model for Effective Fake News Detection with a Novel COVID-19 Dataset
Authors:	Rohit K. Kaliyar, Anurag Goswami and Pratik Narang
Abstract:	Due to the increasing number of users in social media, news articles can be quickly published or share among users without knowing its credibility and authenticity. Fast spreading of fake news articles using different social media platforms can create inestimable harm to society. These actions could seriously jeopardize the reliability of news media platforms. So it is imperative to prevent such fraudulent activities to foster the credibility of such social media platforms. An efficient automated tool is a primary necessity to detect such misleading articles. Considering the issues mentioned earlier, in this paper, we propose a hybrid model using multiple branches of the convolutional neural network (CNN) with Long Short Term Memory (LSTM) layers with different kernel sizes and filters. To make our model deep, which consists of three dense layers to extract more powerful features automatically. In this research, we have created a dataset (FN-COV) collecting 69976 fake and real news articles during the pandemic of COVID-19 with tags like social-distancing, covid19, and quarantine. We have validated the performance of our proposed model with one more real-time fake news dataset: PHEME. The capability of combined kernels and layers of our C-LSTM network is lucrative towards both the datasets. With our proposed model, we achieved an accuracy of 91.88% with PHEME, which is higher as compared to existing models and 98.62% with FN-COV dataset.
Download

Paper Nr:	203
Title:	Planning with Hierarchical Temporal Memory for Deterministic Markov Decision Problem
Authors:	Petr Kuderov and Aleksandr I. Panov
Abstract:	Sequential decision making is among the key problems in Artificial Intelligence. It can be formalized as Markov Decision Process (MDP). One approach to solve it, called model-based Reinforcement Learning (RL), combines learning the model of the environment and the global policy. Having a good model of the environment opens up such properties as data efficiency and targeted exploration. While most of the memory-based approaches are based on using Artificial Neural Networks (ANNs), in our work we instead draw the ideas from Hierarchical Temporal Memory (HTM) framework, which is based on human-like memory model. We utilize it to build an agent’s memory that learns the environment dynamics. We also accompany it with an example of planning algorithm, that enables the agent to solve RL tasks.
Download

Paper Nr:	210
Title:	An Effective Driver Intention and Trajectory Prediction for Autonomous Vehicle based on LSTM
Authors:	Fatimetou El Jili
Abstract:	In order to make the navigation system of autonomous vehicle more robust and safe in urban environment we propose in this paper a model for driver intention prediction and trajectory prediction. The proposed model is based on LSTM (long short term memory). The model was trained on database of features collected from the driving simulator CARLA. This paper treats four type of intentions, turn left, turn right, go straight and stopping intention. Two cases were treated, the first case is to predict intention before it occurs, the second case corresponds to intention recognition, where the driver already starts maneuvering the intention. Both cases are treated by the same model. The model shows better performances for the second case than the first case with small differences. The main strength of our model is that it gives good performances with a small set of features. The accuracy of the model is 96% for intention prediction and 97% for the intention recognition. The proposed method for trajectory prediction reach an accuracy of 99.9%. Those accuracies are higher than what we found in state of art.
Download

Paper Nr:	211
Title:	Tourism Forecast with Weather, Event, and Cross-industry Data
Authors:	Simone Lionetti, Daniel Pfäffli, Marc Pouly, Tim vor der Brück and Philipp Wegelin
Abstract:	The ability to make accurate forecasts on the number of customers is a pre-requisite for efficient planning and use of resources in various industries. It also contributes to global challenges of society such as food waste. Tourism is a domain particularly focussed on short-term forecasting for which the existing literature suggests that calendar and weather data are the most important sources for accurate prediction. We collected and make available a dataset with visitor counts over ten years from four different businesses representative for the tourism sector in Switzerland, along with nearly a thousand features comprising weather, calendar, event and lag information. Evaluation of a plethora of machine learning models revealed that even very advanced deep learning models as well as industry benchmarks show performance at most on a par with simple (piecewise) linear models. Notwithstanding the fact that weather and event features are relevant, contrary to expectations, they proved insufficient for high-quality forecasting. Moreover, and again in contradiction to the existing literature, performance could not be improved by including cross-industry data.
Download

Paper Nr:	212
Title:	Improving Decision-Making-Process for Robot Navigation Under Uncertainty
Authors:	Mohamed I. Khedher, Mallek S. Mziou and Makhlouf Hadji
Abstract:	Designing an autonomous system is a challenging task nowadays, and this is mainly due to two challenges such as conceiving a reliable system in terms of decisions accuracy (performance) and guaranteeing the robustness of the system to noisy inputs. A system is called efficient, if it is simultaneously reliable and robust. In this paper, we consider robot navigation under uncertain environments in which robot sensors may generate disturbed measures affecting the robot decisions. We aim to propose an efficient decision-making model, based on Deep Neural Network (DNN), for robot navigation. Hence, we propose an adversarial training step based on data augmentation to improve robot decisions under uncertain environment. Our contribution is based on investigating data augmentation which is based on uncertainty noise to improve the robustness and performance of the decision model. We also focus on two metrics, Efficiency and Pareto Front, combining robustness and performance to select the best data augmentation rate. In the experiment stage, our approach is validated on a public robotic data-set.
Download

Paper Nr:	214
Title:	Analyzing Adversarial Attacks against Deep Learning for Robot Navigation
Authors:	Mohamed Ibn Khedher and Mehdi Rezzoug
Abstract:	The autonomous system sector continues to experiment and is still progressing every day. Currently, it affects several applications, namely robots, autonomous vehicles, planes, ships, etc. The design of an autonomous system remains a challenge despite all the associated technological development. One of such challenges is the robustness of autonomous system decision in an uncertain environment and their impact on the security of systems, users and people around. In this work, we deal with the navigation of an autonomous robot in a labyrinth room. The objective of this paper is to study the efficiency of a decision-making model, based on Deep Neural Network, for robot navigation. The problem is that, under uncertain environment, robot sensors may generate disturbed measures affecting the robot decisions. The contribution of this work is the proposal of a system validation pipeline allowing the study of its behavior faced to adversarial attacks i.e. attacks consisting in slightly disturbing the input data. In a second step, we investigate the robustness of robot decision-making by applying a defence technique such as adversarial training. In the experiment stage, our study uses a on a public robotic dataset.
Download

Paper Nr:	215
Title:	Dynamic and Scalable Deep Neural Network Verification Algorithm
Authors:	Mohamed I. Khedher, Hatem Ibn-Khedher and Makhlouf Hadji
Abstract:	Deep neural networks have widely used for dealing with complex real-world problems. However, a major concern in applying them to safety-critical systems is the great difficulty in providing formal guarantees about their behavior. Verifying its behavior means study the evolution of its outputs depending on the variation of its inputs. This verification is crucial in an uncertain environment where neural network inputs are noisy. In this paper, we propose an efficient technique for verifying feed-forward neural networks properties. In order to quantify the behavior of the proposed algorithm, we introduce different neural network scenarios to highlight the robustness according to predefined metrics and constraints. The proposed technique is based on the linearization of the non-convex Rectified Linear Unit (ReLU) activation function using the Big-M optimization approach. Moreover, we contribute by an iterative process to find the largest input range verifying (and then defining) the neural network proprieties of neural networks.
Download

Paper Nr:	221
Title:	Identifying Depression Clues using Emotions and AI
Authors:	Ricardo Martins, José J. Almeida, Pedro Henriques and Paulo Novais
Abstract:	According to the World Health Organization (WHO), close to 300 million people of all ages suffer from depression. Also, for WHO, depression is the leading reason for disability worldwide and is a major contributor to the global burden of disease. Different than the mood fluctuation raised by the common life’s activities, depression can be a serious health problem, particularly when it is a long-term and mid/high intensity. Luckily, despite depression is a silent disease, people when suffering leaves some clues. Due to the massive use of social media, these clues can be collected through the texts posted on social media, such as Twitter, Facebook, Instagram, and later, analysed to identify if the writing style matches with a depressive pattern. This paper presents an approach that can be applied by Machine Learning models to help psychologists to identify depressive clues in texts. The model examines profiles on Twitter based on clues provided by users in their posts. Combining Sentiment Analysis, Machine Learning and Natural Language Processing techniques, we achieved a precision of 98% by Machine Learning models when identifying Twitter profiles that post potential depressive texts.
Download

Paper Nr:	223
Title:	Word Reordering and Comma Insertion Integrated with Shift-Reduce Dependency Parsing
Authors:	Kota Miyachi, Tomohiro Ohno and Shigeki Matsubara
Abstract:	Japanese has widely recognized as relatively free word order language. However, since Japanese word order is not completely arbitrary and has some sort of preference, even native Japanese writers often write Japanese sentences that are grammatically well-formed but not easy to read. Furthermore, in Japanese sentences, a comma plays an important role in explicitly separating the constituents such as words and phrases. This paper proposes a method of word reordering and comma insertion for hard-to-read Japanese sentences so that they become easier to read, as basic technique. Our contribution is to show the feasibility of concurrently performing word reordering, comma insertion, and dependency parsing.
Download

Paper Nr:	227
Title:	Multi-feature and Modular Pedestrian Intention Prediction using a Monocular Camera
Authors:	Mostafa Waleed and Amr El Mougy
Abstract:	Accurate prediction of the intention of pedestrians to cross the path of vehicles is highly important to ensure their safety. The accuracy of these intention prediction systems is dependent on the recognition of several pedestrian-related features such as body pose, head pose, pedestrian speed, and passing direction, as well as accurate analysis of the developing traffic situation. Previous research efforts often focus only on a subset of these features, therefore producing inaccurate or incomplete results. Accordingly, this paper presents a comprehensive model for pedestrian intention prediction that incorporates the recognition of all the above features. We also adopt the Constant Velocity Model to estimate the future positions of pedestrians as early as possible. Our model includes a reasoning engine that produces a decision based on the output of the recognition systems of all the aforementioned features. We also consider occlusion scenarios that happen when multiple pedestrians are crossing simultaneously from the same or different directions. Our model is tested on well-known datasets as well as a real autonomous vehicle, and the results show high accuracy in predicting the intention of pedestrians in different scenarios, including ones with occlusion among pedestrians.
Download

Paper Nr:	229
Title:	Neurosymbolic Spike Concept Learner towards Neuromorphic General Intelligence
Authors:	Ahmad N. Wahab, Khaled Mahbub and Abdel-Rahman Tawil
Abstract:	Current research in the area of concept learning makes use of deep learning and ensembles methods to learn concepts. Concept learning allows us to combine heterogeneous entities in data which could collectively identify as individual concepts. Heterogeneity and compositionality are crucial areas to explore in machine learning as it has the potential to contribute profoundly to artificial general intelligence. We investigate the use of spiking neural networks for concept learning. Spiking neurones inclusively model the temporal properties as observed in biological neurones. A benefit of spike-based neurones allows for localised learning rules that only adapts connections between relevant neurones. In this position paper, we propose a technique allowing dynamic formation of synapse (connections) in spiking neural networks, the basis of structural plasticity. Achieving dynamic formation of synapse allows for a unique approach to concept learning with a malleable neural structure. We call this technique Neurosymbolic Spike-Concept Learner (NS-SCL). The limitations of NS-SCL can be overcome with the neuromorphic computing paradigm. Furthermore, introducing NS-SCL as a technique on neuromorphic platforms should motivate a new direction of research towards Neuromorphic General Intelligence (NGI), a term we define to some extent.
Download

Paper Nr:	230
Title:	Masked Hard Coverage Mechanism on Pointer-generator Network for Natural Language Generation
Authors:	Ting Hu and Christoph Meinel
Abstract:	Natural Language Generation (NLG) task is to generate natural language utterances from structured data. Seq2seq-based systems show great potentiality and have been widely explored for NLG. While they achieve good generation performance, over-generation and under-generation issues still arise in the generated results. We propose maintaining a masked hard coverage mechanism in the pointer-generator network, a seq2seq-based architecture that trains a switch policy to produce output sequences by partially copying from input structured data. The proposed mechanism can be regarded as the inner controlling module to keep track of the copying history and force the network to generate sentences accurately covering all information provided in structured data. Experimental results show that our coverage mechanism alleviates the over-generation and under-generation issues and achieves decent performance on the E2E NLG dataset.
Download

Paper Nr:	234
Title:	A Secure Integrated Fog Cloud-IoT Architecture based on Multi-Agents System and Blockchain
Authors:	Chaima Gharbi, Lobna Hsairi and Ezzeddine Zagrouba
Abstract:	Nowadays, the integration of Cloud Computing and the Internet of Things (Cloud-IoT) has drawn attention as new technologies in the Future Internet. Cloud-IoT accommodates good solutions to address real-world problems by offering new services in real-life scenarios. Nonetheless, the traditional Cloud-IoT will be probably not going to give suitable service to the user as it handles enormous amounts of data at a single server. Furthermore, the Cloud-IoT shows huge security and privacy problems that must be solved. To address these issues, we propose an integrated Fog Cloud-IoT architecture based on Multi-Agents System and Blockchain technology. Multi-Agents System has proven itself in decision-making aspects, distributed execution, and its effectiveness in acting in the event of an intrusion without user intervention. On the other side, we propose Blockchain technology as a distributed, public, authentic ledger to record the transactions. The Blockchain represents a great advantage to the next generation computing to ensures data integrity and to allows low latency access to large amounts of data securely. We evaluated the performance of our proposed architecture and compared it with the existing models. The result of our evaluation shows that performance is improved by reducing the response time.
Download

Paper Nr:	255
Title:	Unsupervised Word Sense Disambiguation based on Word Embedding and Collocation
Authors:	Shangzhuang Han and Kiyoaki Shirai
Abstract:	This paper proposes a novel unsupervised word sense disambiguation (WSD) method. It utilizes two useful features for WSD. One is contextual information of a target word. The similarity between words in a context and a sense of a target word is measured based on the pre-trained word embedding, then the most similar sense to the context is chosen. Furthermore, we introduce a procedure not to use irrelevant words in a context in a calculation of the similarity. The other is a collocation, which is an idiomatic phrase including a target word. High-precision rules to determine a sense by a collocation is automatically acquired from a raw corpus. Finally, the above two methods are integrated into our final WSD system. Results of the experiments using Senseval-3 English lexical sample task showed that our proposed method could improve the precision by 4.7 point against the baseline.
Download

Paper Nr:	262
Title:	Is Your Chatbot Perplexing?: Confident Personalized Conversational Agent for Consistent Chit-Chat Dialogue
Authors:	Young Y. Na, Junekyu Park and Kyung-Ah Sohn
Abstract:	Chatbots are being researched and employed not only in academic settings but also in many fields as an application. Ultimately, conversational agents attempt to produce human-like responses along with dialogues. To achieve this goal, we built a novel framework that processes complex data consisting of personalities and utterances and fine-tuned a large-scale self-attention-based language model. We propose a consistent personalized conversational agent(CPC-Agent) for the framework. Our model was designed to utilize the complex knowledge of a dataset to achieve accuracy and consistency. Together with a distractor mechanism, we could generate confident responses. We compared our model to state-of-the-art models using automated metrics. Our model scored 3.29 in perplexity, 17.59 in F1 score, and 79.5 in Hits@1. In particular, the perplexity result was almost four times smaller than that of the current state-of-the-art model that scored 16.42. In addition, we conducted a human evaluation of each model to determine its response quality because the automatic evaluation metrics in dialogue tasks are still considered insufficient. Our model achieved the best rates from the voters, which indicated that our model is adequate for practical use.
Download

Paper Nr:	269
Title:	Multi-objective Classification and Feature Selection of Covid-19 Proteins Sequences using NSGA-II and MAP-Elites
Authors:	Vijay Sambhe, Shanmukha Rajesh, Enrique Naredo, Douglas M. Dias, Meghana Kshirsagar and Conor Ryan
Abstract:	The advent of the Covid-19 pandemic has resulted in a global crisis making the health systems vulnerable, challenging the research community to find novel approaches to facilitate early detection of infections. This open-up a window of opportunity to exploit machine learning and artificial intelligence techniques to address some of the issues related to this disease. In this work, we address the classification of ten SARS-CoV-2 protein sequences related to Covid-19 using k-mer frequency as features and considering two objectives; classification performance and feature selection. The first set of experiments considered the objectives one at the time, four techniques were used for the feature selection and twelve well known machine learning methods, where three are neural network based for the classification. The second set of experiments considered a multi-objective approach where we tested a well known multi-objective approach Non-dominated Sorting Genetic Algorithm II (NSGA-II), and the Multi-dimensional Archive of Phenotypic Elites (MAP-Elites), which considers quality+diversity containers to guide the search through elite solutions. The experimental results shows that ResNet and PCA is the best combination using single objectives. Whereas, for the mulit-classification, NSGA-II outperforms ME with two out of three classifiers, while ME gets competitive results bringing more diverse set of solutions.
Download

Paper Nr:	274
Title:	Comparing Dependency-based Compositional Models with Contextualized Word Embeddings
Authors:	Pablo Gamallo, Manuel P. Corral and Marcos Garcia
Abstract:	In this article, we compare two different strategies to contextualize the meaning of words in a sentence: both distributional models that make use of syntax-based methods following the Principle of Compositionality and Transformer technology such as BERT-like models. As the former methods require controlled syntactic structures, the two approaches are compared against datasets with syntactically fixed sentences, namely subject-predicate and subject-predicate-object expressions. The results show that syntax-based compositional approaches working with syntactic dependencies are competitive with neural-based Transformer models, and could have a greater potential when trained and developed using the same resources.
Download

Paper Nr:	276
Title:	On Informative Tweet Identification for Tracking Mass Events
Authors:	Renato S. João
Abstract:	Twitter has been heavily used as an important channel for communicating and discussing about events in real-time. In such major events, many uninformative tweets are also published rapidly by many users, making it hard to follow the events. In this paper, we address this problem by investigating machine learning methods for automatically identifying informative tweets among those that are relevant to a target event. We examine both traditional approaches with a rich set of handcrafted features and state of the art approaches with automatically learned features. We further propose a hybrid model that leverages both the handcrafted features and the automatically learned ones. Our experiments on several large datasets of real-world events show that the latter approaches significantly outperform the former and our proposed model performs the best, suggesting highly effective mechanisms for tracking mass events.
Download

Paper Nr:	278
Title:	AutoGE: A Tool for Estimation of Grammatical Evolution Models
Authors:	Muhammad S. Ali, Meghana Kshirsagar, Enrique Naredo and Conor Ryan
Abstract:	AutoGE (Automatic Grammatical Evolution), a new tool for the estimation of Grammatical Evolution (GE) parameters, is designed to aid users of GE. The tool comprises a rich suite of algorithms to assist in fine tuning BNF grammar to make it adaptable across a wide range of problems. It primarily facilitates the identification of optimal grammar structures, the choice of function sets to achieve improved or existing fitness at a lower computational overhead over the existing GE setups. This research work discusses and reports initial results with one of the key algorithms in AutoGE, Production Rule Pruning, which employs a simple frequency-based approach for identifying less worthy productions. It captures the relationship between production rules and function sets involved in the problem domain to identify optimal grammar structures. Preliminary studies on a set of fourteen standard Genetic Programming benchmark problems in the symbolic regression domain show that the algorithm removes less useful terminals and production rules resulting in individuals with shorter genome lengths. The results depict that the proposed algorithm identifies the optimal grammar structure for the symbolic regression problem domain to be arity-based grammar. It also establishes that the proposed algorithm results in enhanced fitness for some of the benchmark problems.
Download

Paper Nr:	280
Title:	A Measurement for Essential Conflict in Dempster-Shafer Theory
Authors:	Wenjun Ma, Jieyu Zhan and Yuncheng Jiang
Abstract:	Dempster’s combination rule in Dempster-Shafer (D-S) theory is widely used in data mining, machine learning, clustering and database systems. In these applications, the counter-intuitive result is often obtained with this rule when the combination of mass function is performed without checking whether original beliefs are in conflict. In this paper, a new type of conflict called essential conflict has been revealed with two characteristics: (i) it is an essential factor that leading to the counter-intuitive result by turning a possible state into a necessary true state or an impossible state; (ii) it cannot be corrected by the combination process of any new mass functions. After showing that the existing conflict measurements in D-S theory have the limitations to address the essential conflict and presenting a formalism about the concept of essential conflict, we propose a measurement of essential conflict between two mass functions based on the mass value and the intersection relation of their focal elements. We argue that if there exists a focal element of one mass function, such that the intersection of it and any focal element of another mass function is an empty set, then the essential conflict is caused and Dempster’s combination rule is not applicable.
Download

Paper Nr:	281
Title:	A Landscape Photograph Localisation Method with a Genetic Algorithm using Image Features
Authors:	Hideo Nagashima and Tetsuya Suzuki
Abstract:	It improves the utility value of landscape photographs to identify their shooting locations and shooting directions because geolocated photographs can be used for location-oriented search systems, verification of historically valuable photographs and so on. However, a large amount of labor is required to perform manual shooting location search. Therefore, we are developing a location search system for landscape photographs. To find where and how a given landscape photograph was taken, the system puts virtual cameras in three-dimensional terrain model and adjusts their parameters using a genetic algorithm. The system does not realize efficient search because it has problems such as a long processing time, a multimodal problem and optimization by genetic algorithms. In this research, we propose several efficient search methods using image features and show experimental results for evaluation of them.
Download

Paper Nr:	18
Title:	Solving Maximal Stable Set Problem via Deep Reinforcement Learning
Authors:	Taiyi Wang and Jiahao Shi
Abstract:	This paper provides an innovative method to approximate the optimal solution to the maximal stable set problem, a typical NP-hard combinatorial optimization problem. Different from traditional greedy or heuristic algorithms, we combine graph embedding and DQN-based reinforcement learning to make this NP-hard optimization problem trainable so that the optimal solution over new graphs can be approximated. This appears to be a new approach in solving maximal stable set problem. The learned policy is to choose a sequence of nodes incrementally to construct the stable set, with action determined by the outputs of graph embedding network over current partial solution. Our numerical experiments suggest that the proposed algorithm is promising in tackling the maximum stable independent set problem.
Download

Paper Nr:	24
Title:	Bridging the Technology Gap between Industry and Semantic Web: Generating Databases and Server Code from RDF
Authors:	Markus Schröder, Michael Schulze, Christian Jilek and Andreas Dengel
Abstract:	Despite great advances in the area of Semantic Web, industry rather seldom adopts Semantic Web technologies and their storage and query approaches. Instead, relational databases (RDB) are often deployed to store business-critical data, which are accessed via REST interfaces. Yet, some enterprises would greatly benefit from Semantic Web related datasets which are usually represented with the Resource Description Framework (RDF). To bridge this technology gap, we propose a fully automatic approach that generates suitable RDB models with REST APIs to access them. In our evaluation, generated databases from different RDF datasets are examined and compared. Our findings show that the databases sufficiently reflect their counterparts while the API is able to reproduce rather simple SPARQL queries. Potentials for improvements are identified, for example, the reduction of data redundancies in generated databases.
Download

Paper Nr:	27
Title:	Predicting the Progress of Vehicle Development Projects: An Approach for the Identification of Input Features
Authors:	Oliver Böhme and Tobias Meisen
Abstract:	Today project managers estimate time and other project relevant key performance indicators by using project management tools e.g. milestone trend analysis. We believe that predicting the project’s progress with traditional methods will soon reach its limitations due to the increasing complexity in vehicle development. Machine learning methods provide one possible solution. The vision is to predict the progress of development projects in the early stages of the project. In order to make this vision come true, we need to define measurable input features for machine learning models. In this paper, we focus on representing an approach to identify parameters that exert influence on the progress of development projects.
Download

Paper Nr:	28
Title:	The Person Index Challenge: Extraction of Persons from Messy, Short Texts
Authors:	Markus Schröder, Christian Jilek, Michael Schulze and Andreas Dengel
Abstract:	When persons are mentioned in texts with their first name, last name and/or middle names, there can be a high variation which of their names are used, how their names are ordered and if their names are abbreviated. If multiple persons are mentioned consecutively in very different ways, especially short texts can be perceived as “messy”. Once ambiguous names occur, associations to persons may not be inferred correctly. Despite these eventualities, in this paper we ask how well an unsupervised algorithm can build a person index from short texts. We define a person index as a structured table that distinctly catalogs individuals by their names. First, we give a formal definition of the problem and describe a procedure to generate ground truth data for future evaluations. To give a first solution to this challenge, a baseline approach is implemented. By using our proposed evaluation strategy, we test the performance of the baseline and suggest further improvements. For future research the source code is publicly available.
Download

Paper Nr:	31
Title:	Generating Commonsense Ontologies with Answer Set Programming
Authors:	Stefan Jakob, Alexander Jahl, Harun Baraki and Kurt Geihs
Abstract:	The use of commonsense knowledge is essential for the interaction of humans and robots in a smart environment. This need arises from the way humans naturally communicate with each other, in which most details are usually omitted due to common background knowledge. To enable such communication with a robot, it has to be equipped with a commonsense knowledge representation that supports reasoning. Ontologies could be a suitable approach. However, current ontology frameworks lack dynamic adaptability, are monotonous, are missing negation as failure, and are not designed for huge amounts of data. This paper presents a new way to model ontologies based on a non-monotonic reasoning formalism. Our ontology modelling framework, called ARRANGE, allows for the automatic integration of graph-based knowledge sources to generate ontologies and provides corresponding tools. The presented experiments show the applicability of the generated ontologies and the performance of the ontology generation, the ontology reasoning, and the query resolution.
Download

Paper Nr:	33
Title:	CogToM: A Cognitive Architecture Implementation of the Theory of Mind
Authors:	Fabio Grassiotto and Paula P. Costa
Abstract:	Mind-blindness, a typical trait of autism, is the inability of an individual to attribute mental states to others. This cognitive divergence prevents the proper interpretation of the intentions and the beliefs of other individuals in a given scenario, typically resulting in social interaction problems. In this work, we propose CogToM, a novel cognitive architecture designed to process the output of computer systems and to reason according to the Theory of Mind. In particular, we present a computational implementation for the psychological model of the Theory of Mind proposed by Baron-Cohen and we explore the usefulness of the concepts of Affordances and Intention Detection to augmenting the effectiveness of the proposed architecture. We verify the results by evaluating both a canonical false-belief and a number of the Facebook bAbI dataset tasks.
Download

Paper Nr:	41
Title:	A History-based Framework for Online Continuous Action Ensembles in Deep Reinforcement Learning
Authors:	Renata G. Oliveira and Wouter Caarls
Abstract:	This work seeks optimized techniques of action ensemble deep reinforcement learning to decrease the hyperparameter tuning effort as well as improve performance and robustness, while avoiding parallel environments to make the system applicable to real-world robotic applications. The approach is a history-based framework where different DDPG policies are trained online. The framework’s contributions lie in maintaining a temporal moving average of policy scores, and selecting the actions of the best scoring policies using a single environment. To measure the sensitivity of the ensemble algorithm to the hyperparameter settings, groups were created that mix different amounts of good and bad DDPG parameterizations. The bipedal robot half cheetah environment validated the framework’s best strategy surpassing the baseline by 45%, even with not all good hyperparameters. It presented overall lower variance and superior results with mostly bad parameterization.
Download

Paper Nr:	45
Title:	Incremental Learning for Real-time Partitioning for FPGA Applications
Authors:	Belhedi Wiem, Kammoun Ahmed and Hireche Chabha
Abstract:	The co-design approach consists in defining all the sub-tasks of an application to be integrated and distributed on software or hardware targets. The introduction of conventional cognitive reasoning can solve several problems such as real-time hardware/software classification for FPGA-based applications. However, this requires the availability of large databases, which may conflict with real-time applications. The proposed method is based on the Incremental Kernel SVM (InKSVM) model. InKSVM learns incrementally, as new data becomes available over time, in order to efficiently process large, dynamic data and reduce computation time. As a result, it relaxes the assumption of complete data availability and provides fully autonomous performance. Hence, in this paper, an incremental learning algorithm for hardware/software partitioning is presented. Starting from a real database collected from our FPGA experiments, the proposed approach uses InKSVM to perform the task classification in hardware and software. The proposal has been evaluated in terms of classification efficiency. The performance of the proposed approach was also compared to reference works in the literature. The results of the evaluation consist in empirical evidence of the superiority of the InKSVM over state-of-the- art progressive learning approaches in terms of model accuracy and complexity.
Download

Paper Nr:	67
Title:	A Diagnosis Support System for Veterinary Necropsy based on Bayesian Networks
Authors:	Vianney Sicard, Sébastien Assié, Laëtitia Dorso, Florian Chocteau and Sébastien Picault
Abstract:	Veterinary autopsy requires a high level of expertise and skills that not all veterinarians necessarily master, especially in the context of the desertification of rural areas. The development of support systems is a challenging issue, since such a tool, to be considered relevant and accepted by practitioners in their diagnosis process, must avoid any black box effect. The diagnosis support system we introduce here, IVAN (“Innovative Veterinary Assisted Necropsy”), aims to engage the user in an explicit, understandable, validable and reviewable process, able to cope with the specific issues of cattle necropsy. Besides, it provides uncertainty management to deal with approximate lesion descriptions. IVAN relies on a Bayesian network to infer relevant proposals at each step of the diagnostic process. IVAN was trained on a set of real autopsy cases from autopsy reports, and its performance was assessed using another set of reports. In addition, the tool had to provide results in short response time and be able to run the application on mobile device and web server. In addition to demonstrating the feasibility of the approach, IVAN is a first step towards other support systems in other species and in broader contexts than autopsy.
Download

Paper Nr:	81
Title:	Using Agents and Unsupervised Learning for Counting Objects in Images with Spatial Organization
Authors:	Eliott Jacopin, Naomie Berda, Léa Courteille, William Grison, Lucas Mathieu, Antoine Cornuéjols and Christine Martin
Abstract:	This paper addresses the problem of counting objects from aerial images. Classical approaches either consider the task as a regression problem or view it as a recognition problem of the objects in a sliding window over the images, with, in each case, the need of a lot of labeled images and careful adjustments of the parameters of the learning algorithm. Instead of using a supervised learning approach, the proposed method uses unsupervised learning and an agent-based technique which relies on prior detection of the relationships among objects. The method is demonstrated on the problem of counting plants where it achieves state of the art performance when the objects are well separated and tops the best known performances when the objects overlap. The description of the method underlines its generic nature as it could also be used to count objects organized in a geometric pattern, such as spectators in a performance hall.
Download

Paper Nr:	94
Title:	Training an Agent to Find and Reach an Object in Different Environments using Visual Reinforcement Learning and Transfer Learning
Authors:	Evelyn S. Batista, Wouter Caarls, Leonardo A. Forero and Marco C. Pacheco
Abstract:	This paper consists of a study on deep learning by visual reinforcement for autonomous robots through transfer learning techniques. The simulation environments tested in this study are realistic environments where the challenge of the robot was to learn and transfer knowledge in different contexts, taking advantage of the experience of previous environments in future environments. This type of approach, besides adding knowledge to autonomous robots, reduces the number of training epochs for the algorithm even in complex environments, justifying the use of transfer learning techniques.
Download

Paper Nr:	96
Title:	Decoupling State Representation Methods from Reinforcement Learning in Car Racing
Authors:	Juan M. Montoya, Imant Daunhawer, Julia E. Vogt and Marco Wiering
Abstract:	In the quest for efficient and robust learning methods, combining unsupervised state representation learning and reinforcement learning (RL) could offer advantages for scaling RL algorithms by providing the models with a useful inductive bias. For achieving this, an encoder is trained in an unsupervised manner with two state representation methods, a variational autoencoder and a contrastive estimator. The learned features are then fed to the actor-critic RL algorithm Proximal Policy Optimization (PPO) to learn a policy for playing Open AI’s car racing environment. Hence, such procedure permits to decouple state representations from RL-controllers. For the integration of RL with unsupervised learning, we explore various designs for variational autoencoders and contrastive learning. The proposed method is compared to a deep network trained directly on pixel inputs with PPO. The results show that the proposed method performs slightly worse than directly learning from pixel inputs; however, it has a more stable learning curve, a substantial reduction of the buffer size, and requires optimizing 88% fewer parameters. These results indicate that the use of pre-trained state representations has several benefits for solving RL tasks.
Download

Paper Nr:	99
Title:	Drivable Area Extraction based on Shadow Corrected Images
Authors:	Mohamed Sabry, Mostafa El Hayani, Amr Farag, Slim Abdennadher and Amr El Mougy
Abstract:	Drivable area detection is a complex task that needs to operate efficiently in any environmental condition to ensure wide adoption of autonomous vehicles. In the case of low cost camera-based drivable area detection, the spatial information is required to be uniform as much as possible to ensure the robustness and reliability of the results of any algorithm in most weather and illumination conditions. The general change in illumination and shadow intensities present a significant challenge and can cause major accidents if not considered. Moreover, drivable area detection in unstructured environments is more complex due to the absence of vital spatial information such as road markings and lanes. In this paper, a shadow reduction approach combining Computer Vision (CV) - Image Processing (IM) with Deep Learning (DL) is used on a low cost monocular camera based system for reliable and uniform shadow removal. In addition, a validation test is applied with a DL model to validate the approach. This system is developed for the Self-driving Car (SDC) lab at the German University in Cairo (GUC) and is to be used in the shell eco-marathon autonomous competition 2021.
Download

Paper Nr:	106
Title:	Prediction of Cotton Field on Integrated Environmental Data
Authors:	Sarthak Mishra, Long Ma and Nischal Aryal
Abstract:	The agriculture and farming industry plays a vital role in the economy. However, the importance of agriculture cannot be fully quantified in terms of its economic profit. Agriculture affecting global hunger is a much more sensitive and vital topic. One of the leading reasons for this is un-improvised crop production. Crop production is affected by various factors, and monitoring those factors is the key to solving the problem. This paper describes a comprehensive experiment predicting the cotton yield under various environments, such as Acres Harvested, Acres Planted, Soil pH, Bulk Density, Clay-High, Clay-Low, Organic-Carbon, and Water-Area.
Download

Paper Nr:	110
Title:	Obsolescence Prediction based on Joint Feature Selection and Machine Learning Techniques
Authors:	Imen Trabelsi, Besma Zeddini, Marc Zolghadri, Maher Barkallah and Mohamed Haddar
Abstract:	Obsolescence is a serious phenomenon that affects all systems. To reduce its impacts, a well-structured management method is essential. In the field of obsolescence management, there is a great need for a method to predict the occurrence of obsolescence. This article reviews obsolescence forecasting methodologies and presents an obsolescence prediction methodology based on machine learning. The model developed is based on joint a machine learning (ML) technique and feature selection. A feature selection method is applied to reduce the number of inputs used to train the ML technique. A comparative study of the different methods of feature selection is established in order to find the best in terms of precision. The proposed method is tested by simulation on models of mobile phones. Consequently, the use of features selection method in conjunction with ML algorithm surpasses the use of ML algorithm alone.
Download

Paper Nr:	127
Title:	Headway and Following Distance Estimation using a Monocular Camera and Deep Learning
Authors:	Zakaria Charouh, Amal Ezzouhri, Mounir Ghogho and Zouhair Guennoun
Abstract:	We propose a system for monitoring the headway and following distance using a roadside camera and deep learning-based computer vision techniques. The system is composed of a vehicle detector and tracker, a speed estimator and a headway estimator. Both motion-based and appearance-based methods for vehicle detection are investigated. Appearance-based methods using convolutional neural networks are found to be most appropriate given the high detection accuracy requirements of the system. Headway estimation is then carried out using the detected vehicles on a video sequence. The following distance estimation is carried out using the headway and speed estimations. We also propose methods to assess the performance of the headway and speed estimation processes. The proposed monitoring system has been applied to data that we have collected using a roadside camera. The root mean square error of the headway estimation is found to be around 0.045 seconds.
Download

Paper Nr:	131
Title:	A New Benchmark for NLP in Social Sciences: Evaluating the Usefulness of Pre-trained Language Models for Classifying Open-ended Survey Responses
Authors:	Maximilian Meidinger and Matthias Aßenmacher
Abstract:	In order to evaluate transfer learning models for Natural Language Processing on a common ground, numerous general domain (sets of) benchmark data sets have been established throughout the last couple of years. Primarily, the proposed tasks are classification (binary, multi-class), regression or language generation. However, no benchmark data set for (extreme) multi-label classification relying on full-text inputs has been proposed in the area of social science survey research to this date. This constitutes an important gap, as a common data set for algorithm development in this field could lead to more reproducible, sustainable research. Thus, we provide a transparent and fully reproducible preparation of the 2008 American National Election Study (ANES) data set, which can be used for benchmark comparisons of different NLP models on the task of multi-label classification. In contrast to other data sets, our data set comprises full-text inputs instead of bag-of-words representations or similar. Furthermore, we provide baseline performances of simple logistic regression models as well as performance values for recently established transfer learning architectures, namely BERT (Devlin et al., 2018), RoBERTa (Liu et al., 2019) and XLNet (Yang et al., 2019).
Download

Paper Nr:	145
Title:	GREE-COCO: Green Artificial Intelligence Powered Cost Pricing Models for Congestion Control
Authors:	Meghana Kshirsagar, Tanishq More, Rutuja Lahoti, Shreya Adgaonkar, Shruti Jain, Conor Ryan and Vivek Kshirsagar
Abstract:	The objective of the proposed research is to design a system called Green Artificial Intelligence Powered Cost Pricing Models for Congestion Control (GREE-COCO) for road vehicles that address the issue of congestion control through the concept of cost pricing. The motivation is to facilitate smooth traffic flow among densely congested roads by incorporating static and dynamic cost pricing models. The other objective behind the study is to reduce pollution and fuel consumption and encourage people towards positive usage of the public transport system (e.g., bus, train, metro, and tram). The system will be implemented by charging the vehicles driven on a particular congested road during a specific time. The pricing will differ according to the location, type of vehicle, and vehicle count. The cost pricing model incorporates an incentive approach for rewarding the usage of electric/non-fuel vehicles. The system will be tested with analytics gathered from cameras installed for testing purposes in some of the Indian and Irish cities. One of the challenges that will be addressed is to develop sustainable and energy-efficient Artificial Intelligence (AI) models that use less power consumption which results in low carbon emission. The GREE-COCO model consists of three modules: vehicle detection and classification, license plate recognition, and cost pricing model. The AI models for vehicle detection and classification are implemented with You Only Look Once (YOLO) v3, Faster-Region based Convolutional Neural Network (F-RCNN), and Mask-Region based Convolutional Neural Network (Mask RCNN). The selection of the best model depends upon their performance concerning accuracy and energy efficiency. The dynamic cost pricing model is tested with both the Support Vector Machine (SVM) classifier and the Generalised Linear Regression Model (GLM). The experiments are carried out on a custom-made video dataset of 103 videos of different time duration. The initial results obtained from the experimental study indicate that YOLOv3 is best suited for the system as it has the highest accuracy and is more energy-efficient.
Download

Paper Nr:	152
Title:	Understand Watchdogs: Discover How Game Bot Get Discovered
Authors:	Eunji Park, Kyung H. Park and Huy K. Kim
Abstract:	The game industry has long been troubled by malicious activities utilizing game bots. The game bots disturb other game players and destroy the environmental system of the games. For these reasons, the game industry put their best efforts to detect the game bots among players’ characters using the learning-based detections. However, one problem with the detection methodologies is that they do not provide rational explanations about their decisions. To resolve this problem, in this work, we investigate the explainabilities of the game bot detection. We develop the XAI model using a dataset from the Korean MMORPG, AION, which includes game logs of human players and game bots. More than one classification model has been applied to the dataset to be analyzed by applying interpretable models. This provides us explanations about the game bots’ behavior, and the truthfulness of the explanations has been evaluated. Besides, interpretability contributes to minimizing false detection, which imposes unfair restrictions on human players.
Download

Paper Nr:	155
Title:	Domain Adaptation Multi-task Deep Neural Network for Mitigating Unintended Bias in Toxic Language Detection
Authors:	Farshid Faal, Jia Y. Yu and Ketra Schmitt
Abstract:	As online communities have grown, so has the ability to exchange ideas, which includes an increase in the spread of toxic language, including racism, sexual harassment, and other negative behaviors that are not tolerated in polite society. Hence, toxic language detection within online conversations has become an essential application of natural language processing. In recent years, machine learning approaches for toxic language detection have primarily focused on many researchers in academics and industries. However, in many of these machine learning models, non-toxic comments containing specific identity terms, such as gay, Black, Muslim, and Jewish, were given unreasonably high toxicity scores. In this research, we propose a new approach based on the domain adaptation language model and multi-task deep neural network to identify and mitigate this form of unintended model bias in online conversations. We use six toxic language detection and identification tasks to train the model to detect toxic contents and mitigate unintended bias in model prediction. We evaluate our model and compare it with other state-of-the-art deep learning models using specific performance metrics to measure the model bias. In detailed experiments, we show our approach can identify the toxic language in conversations with considerably more robustness to model bias towards commonly-attacked identity groups presented in online conversations in social media.
Download

Paper Nr:	157
Title:	Pairwise Cosine Similarity of Emission Probability Matrix as an Indicator of Prediction Accuracy of the Viterbi Algorithm
Authors:	Guantao Zhao, Ziqiu Zhu, Yinan Sun and Amrinder Arora
Abstract:	The Viterbi Algorithm is the main algorithm for the Most Likely Explanation (MLE) used in the HMM. We study the hypothesis that the prediction accuracy of the Viterbi algorithm can be estimated a priori by computing the arithmetic mean of the cosines of the emission probabilities. Our analysis and experimental results suggest a close relationship between these two quantities.
Download

Paper Nr:	165
Title:	On the Prediction of a Nonstationary Bernoulli Distribution based on Bayes Decision Theory
Authors:	Daiki Koizumi
Abstract:	A class of nonstationary Bernoulli distribution is considered in terms of Bayes decision theory. In this nonstationary class, the Bernoulli distribution parameter follows a random walking rule. Even if this general class is assumed, it is proved that the posterior distribution of the parameter can be obtained analytically with a known hyper parameter. With this theorem, the Bayes optimal prediction algorithm is proposed assuming the 0-1 loss function. Using real binary data, the predictive performance of the proposed model is evaluated comparing to that of a stationary Bernoulli model.
Download

Paper Nr:	188
Title:	LongiControl: A Reinforcement Learning Environment for Longitudinal Vehicle Control
Authors:	Jan Dohmen, Roman Liessner, Christoph Friebel and Bernard Bäker
Abstract:	Reinforcement Learning (RL) might be very promising for solving a variety of challenges in the field of autonomous driving due to its ability to find long-term oriented solutions in complex decision scenarios. For training and validation of a RL algorithm, a simulative environment is advantageous due to risk reduction and saving of resources. This contribution presents an RL environment designed for the optimization of longitudinal control. The focus is on providing an illustrative and comprehensible example for a continuous real-world problem. The environment will be published following the OpenAI Gym interface, allowing for easy testing and comparing of novel RL algorithms. In addition to details on implementation reference is also made to areas where research is required.
Download

Paper Nr:	196
Title:	How Are the Members of a Parliament Arguing? Analysis of an Argument Corpus
Authors:	Mare Koit
Abstract:	Estonian argument corpus includes verbatim records (in Estonian) of sessions held in the Parliament of Estonia (Riigikogu). Arguments used in negotiation and inter-argument relations are annotated in the corpus. Every argument consists of one or more premises, and a claim. By using the corpus, inter-argument relations (rebuttal, attack, and support), argument diagramming (argument structures – basic, convergent, serial, divergent, and linked), and the linguistic features of the arguments are studied. Some problems are discussed in relation to the arguments the members of Riigikogu use when negotiating. Our aim is to add an additional layer to our argument corpus by annotating the structures of arguments as well as extending the corpus in order to make it possible the automatic recognition of arguments in Estonian political texts. A further challenge will be the comparison of discussions in Riigikogu with other parliaments and other languages.
Download

Paper Nr:	204
Title:	Cycle4Value: A Blockchain-based Reward System to Promote Cycling and Reduce CO2 Footprint
Authors:	Alexander K. Seewald, Mihai Ghete, Thomas Wernbacher, Mario Platzer, Josefine Schneider, Dietmar Hofer and Alexander Pfeiffer
Abstract:	In Cycle4Value (C4V), a transparent and low-threshold reward model to promote cycling based on the key technology blockchain is being researched and tested in practice for the first time. The economic, health and ecological benefits are presented in a simple and comprehensible way and, after a plausibility check using a pretrained machine learning model, are converted into a real value, i.e. a cycle token. These units of value are stored in a digital wallet and can be reimbursed in a marketplace set up for testing. The research project goes beyond conventional incentive systems, since 1) the storage of the value units as well as the payment process is decentralised, tamper-proof and transparent, and 2) the real economic and environmental benefit of active cycling is monetized in a fair manner. Initially we describe the background of the project. The main part of this paper concerns ongoing work on the plausibility check which also needs to be able to detect cheating.
Download

Paper Nr:	220
Title:	Leveraging Automated Machine Learning for Text Classification: Evaluation of AutoML Tools and Comparison with Human Performance
Authors:	Matthias Blohm, Marc Hanussek and Maximilien Kintz
Abstract:	Recently, Automated Machine Learning (AutoML) has registered increasing success with respect to tabular data. However, the question arises whether AutoML can also be applied effectively to text classification tasks. This work compares four AutoML tools on 13 different popular datasets, including Kaggle competitions, and opposes human performance. The results show that the AutoML tools perform better than the machine learning community in 4 out of 13 tasks and that two stand out.
Download

Paper Nr:	225
Title:	The Determinants of Social Media Engagement for Fashion Industry in Oman: A Descriptive Analysis
Authors:	Fatma S. Al Rabaani and Aiman M. Said
Abstract:	The use of social media has completely remodelled the way people interact, communicate, and engage. Social media platforms play an essential role in reshaping the relationship between customers and companies. Present companies establish their accounts in social media to reach and engage with their customers, listen and take their opinion, enhance the purchase decision, and increase the revenue. The main goal of this study is to determine the factors that affect customer engagement. From 296 Instagram business accounts with 530,366 posts published, the dataset was scraped and used to understand what impacts customer engagement. Different descriptive analysis techniques were adopted to answer the questions of the study. Among the key finding of this study, customer engagement is positively affected by the number of comments and shares. The number of likes of published posts is not influenced. Moreover, video posts attract more customer interaction than other types of posts. Uncovered the property of three kinds of customer engagement (low, moderately, and high active).
Download

Paper Nr:	237
Title:	Bet-based Evolutionary Algorithms: Self-improving Dynamics in Offspring Generation
Authors:	Simon Reichhuber and Sven Tomforde
Abstract:	Evolutionary Algorithms (EA) are a well-studied field in nature-inspired optimisation. Their success over the last decades has led to a large number of extensions, which are particularly suitable for certain characteristics of specific problems. Alternatively, variants of the basic approach have been proposed, for example to increase efficiency. In this paper, we focus on the latter: We propose to enrich the evolutionary problem with a self- controlling betting strategy to optimise the evolution of individuals over successive generations. For this purpose, each individual is given a betting parameter to be co-optimised, which allows him to improve his chances of “survival” by betting. We analyse the behaviour of our approach compared to standard procedures by using a reference set of complex functional problems.
Download

Paper Nr:	251
Title:	On the Limits to Multi-modal Popularity Prediction on Instagram: A New Robust, Efficient and Explainable Baseline
Authors:	Christoffer Riis, Damian K. Kowalczyk and Lars K. Hansen
Abstract:	Our global population contributes visual content on platforms like Instagram, attempting to express themselves and engage their audiences, at an unprecedented and increasing rate. In this paper, we revisit the popularity prediction on Instagram. We present a robust, efficient, and explainable baseline for population-based popularity prediction, achieving strong ranking performance. We employ the latest methods in computer vision to maximise the information extracted from the visual modality. We use transfer learning to extract visual semantics such as concepts, scenes, and objects, allowing a new level of scrutiny in an extensive, explainable ablation study. We inform feature selection towards a robust and scalable model, but also illustrate feature interactions, offering new directions for further inquiry in computational social science. Our strongest models inform a lower limit to population-based predictability of popularity on Instagram. The models are immediately applicable to social media monitoring and influencer identification.
Download

Paper Nr:	253
Title:	Using Machine Learning to Forecast Air and Water Quality
Authors:	Carolina Silva, Bruno Fernandes, Pedro Oliveira and Paulo Novais
Abstract:	Environmental sustainability is one of the biggest concerns nowadays. With increasingly latent negative impacts, it is substantiated that future generations may be compromised. The research here presented addresses this topic, focusing on air quality and atmospheric pollution, in particular the Ultraviolet index and Carbon Monoxide air concentration, as well as water issues regarding Wastewater Treatment Plants, in particular the pH of water. A set of Machine Learning regressors and classifiers are conceived, tuned, and evaluated in regard to their ability to forecast several parameters of interest. The experimented models include Decision Trees, Random Forests, Multilayer Perceptrons, and Long Short-Term Memory networks. The obtained results assert the strong ability of LSTMs to forecast air pollutants, with all models presenting similar results when the subject was the pH of water.
Download

Paper Nr:	263
Title:	SLPRNet: A 6D Object Pose Regression Network by Sample Learning
Authors:	Zheng Zhang, Xingru Zhou and Houde Liu
Abstract:	Visual grasping holds important implications for robot manipulation situations. As a core procedure in such grasping tasks, pose regression has attracted lots of research attention, among which point cloud based deep learning methods achieve relatively better result. The usual backbone of such network architectures includes sampling, grouping and feature extracting processes. We argue that common sampling techniques like Farthest Point Sampling(FPS), Random Sampling(RS) and Geometry Sampling(GS) hold potential defectiveness. So we devise a pre-posed network which aims at learning to sample the most suitable points in the whole point cloud for a downstream pose regression task and show its superiority comparing to the above-mentioned sampling methods. In conclusion, we propose a Sample Learning Pose Regression network (SLPRNet) to regress each instances pose in a standard grasping situation. Meanwhile, we build a point cloud dataset to train and test our network. In experiment, we reach an average precision(AP) up to 89.8% on dataset generated from Silane and an average distance(ADD) up to 91.0% on YCB. Real-world grasp experiments also verify the validity of our work.
Download

Paper Nr:	271
Title:	Computer Modeling of the Dynamics of Interregional Freight Transport Depending on Macroeconomic Indicators
Authors:	Maksim Tatarintsev, Petr Nikitin and Sergey Korchagin
Abstract:	The purpose of this research is to consider various macroeconomic indicators of world regions(North America, South America, Europe, Mediterranean, Persian Gulf, CIS, South Africa, Central America, West Africa, East Africa, East Asia, South Asia, Southeast Asia, Oceania, Australia) and federal districts in the Russian Federation (Central, North-western, Southern, North Caucasian, Volga, Ural, Siberian, Far Eastern), visualize the received data, determine their impact on the volume of freight traffic between the designated zones, identify patterns with the help of which we could carry out various kinds of forecasting. Further, it is necessary to carry out direct forecasting for several products of the selected industry, applying the necessary math model.
Download

Area 2 - Agents

Full Papers

Paper Nr:	13
Title:	Reproducing Evacuation Behaviors of Evacuees during the Great East Japan Earthquake using the Evacuation Decision Model with Realistic Settings
Authors:	Akira Tsurushima
Abstract:	The analysis of evacuation behaviors from the video captured during the Great East Japan Earthquake revealed that the evacuation behaviors of fleeing and dropping down were affected by the distance from the exits. These behaviors were reproduced through simulations by employing the evacuation decision model, which is a model of herd behaviors during evacuations; this showed that these unique evacuation behaviors could be reproduced by simple herd behaviors. However, the results are questionable owing to the oversimplified settings of the simulations, such as the different number and density of agents and the overlooked physical constraints. We conduct simulations with settings that are more representative of those in the video clip. The unique evacuation behavior is also reproduced with our simulation setting but for limited ranges of parameter values. The analysis of the results reveals that the parameters related to the vicinity of an agent are significant; this lead to the hypothesis that the attention of evacuees is narrowed to 20 degrees with a relatively long range during evacuations.
Download

Paper Nr:	29
Title:	SAT-MARL: Specification Aware Training in Multi-Agent Reinforcement Learning
Authors:	Fabian Ritz, Thomy Phan, Robert Müller, Thomas Gabor, Andreas Sedlmeier, Marc Zeller, Jan Wieghardt, Reiner Schmid, Horst Sauer, Cornel Klein and Claudia Linnhoff-Popien
Abstract:	A characteristic of reinforcement learning is the ability to develop unforeseen strategies when solving problems. While such strategies sometimes yield superior performance, they may also result in undesired or even dangerous behavior. In industrial scenarios, a system’s behavior also needs to be predictable and lie within defined ranges. To enable the agents to learn (how) to align with a given specification, this paper proposes to explicitly transfer functional and non-functional requirements into shaped rewards. Experiments are carried out on the smart factory, a multi-agent environment modeling an industrial lot-size-one production facility, with up to eight agents and different multi-agent reinforcement learning algorithms. Results indicate that compliance with functional and non-functional constraints can be achieved by the proposed approach.
Download

Paper Nr:	44
Title:	Indoor Point-to-Point Navigation with Deep Reinforcement Learning and Ultra-Wideband
Authors:	Enrico Sutera, Vittorio Mazzia, Francesco Salvetti, Giovanni Fantin and Marcello Chiaberge
Abstract:	Indoor autonomous navigation requires a precise and accurate localization system able to guide robots through cluttered, unstructured and dynamic environments. Ultra-wideband (UWB) technology, as an indoor positioning system, offers precise localization and tracking, but moving obstacles and non-line-of-sight occurrences can generate noisy and unreliable signals. That, combined with sensors noise, unmodeled dynamics and environment changes can result in a failure of the guidance algorithm of the robot. We demonstrate how a power-efficient and low computational cost point-to-point local planner, learnt with deep reinforcement learning (RL), combined with UWB localization technology can constitute a robust and resilient to noise short-range guidance system complete solution. We trained the RL agent on a simulated environment that encapsulates the robot dynamics and task constraints and then, we tested the learnt point-to-point navigation policies in a real setting with more than two-hundred experimental evaluations using UWB localization. Our results show that the computational efficient end-to-end policy learnt in plain simulation, that directly maps low-range sensors signals to robot controls, deployed in combination with ultra-wideband noisy localization in a real environment, can provide a robust, scalable and at-the-edge low-cost navigation system solution.
Download

Paper Nr:	60
Title:	Adaptive Planning Method for Operations of a Multi-satellite Swarm for Earth Remote Sensing in Real Time
Authors:	Petr Skobelev, Elena Simonova, Vladimir Galuzin, Anastasiya Galitskaya and Vitaly Travin
Abstract:	The paper describes a method for adaptive planning of imaging operations for a multi-satellite swarm in real time, based on a multi-agent approach. The key object in this approach is the intelligent agent of an application for imaging of the observation object. Its goal is the most advantageous placement in the schedule. The solution to the optimization problem is obtained as a result of reaching an equilibrium point in multiple negotiations between agents through mutual compromises and concessions. The paper provides a brief problem statement of planning the operation of a multi-satellite swarm for Earth remote sensing (ERS). Furthermore, it describes the developed method, which makes it possible to process applications for imaging observation objects in real time. The paper also presents results of experimental studies that demonstrate efficiency of the developed multi-agent method in solving this problem versus traditional approaches. Finally, prospects for further development and practical application of the presented method are discussed.
Download

Paper Nr:	73
Title:	A Multi-Agent Negotiation Strategy for Reducing the Flowtime
Authors:	Ellie Beauprez, Anne-Cécile Caron, Maxime Morge and Jean-Christophe Routier
Abstract:	In this paper, we study the problem of task reallocation for load-balancing in distributed data processing models that tackle vast amount of data. In this context, we propose a novel strategy based on cooperative agents used to optimize the rescheduling of tasks for multiple jobs submitted by users in order to be executed as soon as possible. It allows an agent to determine locally the next task to process and the next task to delegate according to its knowledge, its own belief base and its peer modelling. The novelty of our strategy lies in the ability of agents to identify opportunities and bottleneck agents, and afterwards to reallocate some of the tasks. Our contribution is that, thanks to concurrent bilateral negotiations, tasks are continuously reallocated according to the local perception and the peer modelling of agents. In order to evaluate the responsiveness of our approach, we implement a prototype testbed and our experimentation reveals that our strategy reaches a flowtime which is close to the one reached by the classical heuristic approach and significantly reduces the rescheduling time.
Download

Paper Nr:	105
Title:	A Systematic Review about Requirements Engineering Processes for Multi-Agent Systems
Authors:	Giovane D. Mendonça, Iderli S. Souza Filho and Gilleanes A. Guedes
Abstract:	Requirements engineering is a crucial phase for the software development process, including multi-agent systems. This particular kind of software is composed by agents, autonomous and pro-active entities which can collaborate among themselves to achieve a given goal. However, multi-agent systems have some particular requirements that are not normally found in other software. Taking this into consideration, this paper aims to determine the actual state of the development processes which support requirements engineering for multi-agent systems by means of a systematic review, highlighting the requirements engineering coverage and its support to the BDI model.
Download

Paper Nr:	125
Title:	Multi-Agent Goal Reasoning with the CLIPS Executive in the RoboCup Logistics League
Authors:	Till Hofmann, Tarik Viehmann, Mostafa Gomaa, Daniel Habering, Tim Niemueller and Gerhard Lakemeyer
Abstract:	Production processes in smart factories moved away from a process-centered paradigm into a modular production paradigm, facing the variations in demanded product configurations and deadlines with a flexible production. The RoboCup Logistics League (RCLL) is a robotics competition in the context of in-factory logistics, in which a team of three autonomous mobile robots manufacture dynamically ordered products. The main challenges include task reasoning, multi-agent coordination, and robust execution in a dynamic environment. We present a multi-agent goal reasoning approach where agents continuously reason about which objectives to pursue rather than only planning for a fixed objective. We describe an incremental, distributed formulation of the RCLL problem implemented in the goal reasoning system CLIPS Executive. We elaborate what kind of goals we use in the RCLL, how we use goal trees to define an effective production strategy and how agents coordinate effectively by means of primitive lock actions as well as goal-level resource allocation. The system utilizes a PDDL model to describe domain predicates and actions, as well as to determine the executability and effects of actions during execution. Our agent is able to react to unexpected events, such as a broken machine or a failed action, by monitoring the execution of the plan, re-evaluating goals, and taking over goals which were previously pursued by another robot. We present a detailed evaluation of the system used on real robots.
Download

Paper Nr:	139
Title:	Let’s Do the Time Warp Again: Human Action Assistance for Reinforcement Learning Agents
Authors:	Carter B. Burn, Frederick L. Crabbe and Rebecca Hwa
Abstract:	Reinforcement learning (RL) agents may take a long time to learn a policy for a complex task. One way to help the agent to convergence on a policy faster is by offering it some form of assistance from a teacher who already has some expertise on the same task. The teacher can be either a human or another computer agent, and they can provide assistance by controlling the reward, action selection, or state definition that the agent views. However, some forms of assistance might come more naturally from a human teacher than a computer teacher and vice versa. For instance, a challenge for human teachers in providing action selection is that because computers and human operate at different speed increments, it is difficult to translate what constitutes an action selection for a particular state in a human’s perception to that of the computer agent. In this paper, we introduce a system called Time Warp that allows a human teacher to provide action selection assistance to the agent during critical moments of the training for the RL agent. We find that Time Warp is able to help the agent develop a better policy in less time than an RL agent with no assistance and rivals the performance of computer teaching agents. Time Warp also is able to reach the results with only ten minutes of human training time.
Download

Paper Nr:	192
Title:	Distributed Service Area Control for Ride Sharing by using Multi-Agent Deep Reinforcement Learning
Authors:	Naoki Yoshida, Itsuki Noda and Toshiharu Sugawara
Abstract:	We propose a decentralized system to determine where ride-sharing vehicle agents should wait for passengers using multi-agent deep reinforcement learning. Although numerous drivers have begun participating in ride-sharing services as the demand for these services has increased, much of their time is idle. The result is not only inefficiency but also wasted energy and increased traffic congestion in metropolitan area, while also causing a shortage of ride-sharing vehicles in the surrounding areas. We therefore developed the distributed service area adaptation method for ride sharing (dSAAMS) to decide the areas where each agent should wait for passengers through deep reinforcement learning based on the networks of individual agents and the demand prediction data provided by an external system. We evaluated the performance and characteristics of our proposed method in a simulated environment with varied demand occurrence patterns and by using actual data obtained in the Manhattan area. We compare the performance of our method to that of other conventional methods and the centralized version of the dSAAMS. Our experiments indicate that by using the dSAAMS, agents individually wait and move more effectively around their service territory, provide better quality service, and exhibit better performance in dynamically changing environments than when using the comparison methods.
Download

Paper Nr:	201
Title:	Mixed Deep Reinforcement Learning-behavior Tree for Intelligent Agents Design
Authors:	Lei Li, Lei Wang, Yuanzhi Li and Jie Sheng
Abstract:	Intelligent agent design has increasingly enjoyed the great advancements in real-world applications but most agents are also required to possess the capacities of learning and adapt to complicated environments. In this work, we investigate a general and extendable model of mixed behavior tree (MDRL-BT) upon the option framework where the hierarchical architecture simultaneously involves different deep reinforcement learning nodes and normal BT nodes. The emphasis of this improved model lies in the combination of neural network learning and restrictive behavior framework without conflicts. Moreover, the collaborative nature of two aspects can bring the benefits of expected intelligence, scalable behaviors and flexible strategies for agents. Afterwards, we enable the execution of the model and search for the general construction pattern by focusing on popular deep RL algorithms, PPO and SAC. Experimental performances in both Unity 2D and 3D environments demonstrate the feasibility and practicality of MDRL-BT by comparison with the-state-of-art models. Furthermore, we embed the curiosity mechanism into the MDRL-BT to facilitate the extensions.
Download

Paper Nr:	207
Title:	Automatic Definition of MOISE Organizations for Adaptive Workflows
Authors:	Massimo Cossentino, Salvatore Lopes and Luca Sabatucci
Abstract:	The enactment of dynamic workflows may take advantage of the multi-agent system paradigm. The approach presented in this paper allows exploiting a high-level BPMN process definition to generate an agent organisation that can enact the workflow using different strategies. These are implemented as organisational schemes representing alternative goal decomposition trees. The availability of several equivalent solutions enables the optimisation and adaptation features of the approach. The mapping of the initial workflow to organisations starts with the automatic generation of goals from the BPMN, and it exploits a metamodeling approach to generate MOISE organisation definition.
Download

Paper Nr:	226
Title:	Exploring Narrative Economics: An Agent-based-modeling Platform that Integrates Automated Traders with Opinion Dynamics
Authors:	Kenneth Lomas and Dave Cliff
Abstract:	In seeking to explain aspects of real-world economies that defy easy understanding when analysed via conventional means, Nobel Laureate Robert Shiller has since 2017 introduced and developed the idea of Narrative Economics, where observable economic factors such as the dynamics of prices in asset markets are explained largely as a consequence of the narratives (i.e., the stories) heard, told, and believed by participants in those markets. Shiller argues that otherwise irrational and difficult-to-explain behaviors, such as investors participating in highly volatile cryptocurrency markets, are best explained and understood in narrative terms: people invest because they believe, because they have a heartfelt opinion, about the future prospects of the asset, and they tell to themselves and others stories (narratives) about those beliefs and opinions. In this paper we describe what is, to the best of our knowledge, the first ever agent-based modelling platform that allows for the study of issues in narrative economics. We have created this by integrating and synthesizing research in two previously separate fields: opinion dynamics (OD), and agent-based computational economics (ACE) in the form of minimally-intelligent trader-agents operating in accurately modelled financial markets. We show here for the first time how long-established models in OD and in ACE can be brought together to enable the experimental study of issues in narrative economics, and we present initial results from our system. The program-code for our simulation platform has been released as freely-available open-source software on GitHub, to enable other researchers to replicate and extend our work.
Download

Paper Nr:	242
Title:	A Study on Negotiation for Revealed Information with Decentralized Asymmetric Multi-objective Constraint Optimization
Authors:	Toshihiro Matsui
Abstract:	The control of revealed information from agents is an important issue in cooperative problem solving and negotiation in multi-agent systems. Research in automated negotiation agents and distributed constraint optimization problems address the privacy of agents. While several studies employ the secure computation that completely inhibits to access the information in solution process, a part of the information is necessary for selfish agents to understand the situation of an agreement. In this study, we address a decentralized framework where agents iteratively negotiate and gradually publish the information of their utilities that are employed to determine a solution of a constraint optimization problem among the agents. A benefit of the approach based on constraint optimization problems is its ability of formalization for general problems. We represent both problems to determine newly published information of utility values and to determine a solution based on published utility values as decentralized asymmetric multi-objective constraint optimization problems. As the first study, we investigate the opportunities to design constraints that define simple strategies of agents to control the utility values to be published. For the objectives of individual agents, we also investigate the influence of several social welfare functions. We experimentally show the effect and influence of the heuristics of different criteria to select the published information of agents.
Download

Paper Nr:	273
Title:	Application of Multiagent System and Tabu Search for Truck Dispatching in Open-pit Mines
Authors:	Gabriel Icarte Ahumada and Otthein Herzog
Abstract:	An important and complex process in the mining industry is the material handling process. In this process, trucks must transport materials extracted by shovels to different places at the mine. To enable efficient material handling processes, the decision on the destination of a truck is crucial. Currently, this process is supported by an approach based on centralized systems that apply dispatching criteria. A disadvantage of this approach is not providing a precise dispatching solution because of missing knowledge about potentially changed external conditions and the dependency on a central node. We previously developed a multiagent system (MAS-TD) to solve this problem. In the MAS-TD, intelligent agents that represent real-world equipment interact with each other to generate schedules. In this paper, we evaluate the MAS-TD by comparing it against a Tabu Search procedure. In the evaluation, simulated scenarios based on actual data from a Chilean open-pit mine were used. The results show that both MAS-TD and the Tabu Search procedure are suitable methods to solve the truck dispatching problem in open-pit mines. However, the schedules generated by MAS-TD are more efficient than the schedules generated by the Tabu Search.
Download

Short Papers

Paper Nr:	54
Title:	CAMP-IRL Agents: Extracting Multiple Behavior Profiles for Pedestrian Simulation
Authors:	Nahum Alvarez and Itsuki Noda
Abstract:	Crowd simulation has been subject of study due to its applications in the fields of evacuation management, smart town planning and business strategic placing. Simulations of human behavior have many useful applications but are limited in their flexibility. A possible solution to that issue is to use semi-supervised machine learning techniques to extract action patterns for the simulation. In this paper, we present a model for agent-based crowd simulation that generates agents capable of navigating efficiently across a map attending to different goal driven behaviors. We designed an agent model capable of using the different behavior patterns obtained from training data, imitating the behavior of the real pedestrians and we compared it with other models attending to behavioral metrics.
Download

Paper Nr:	68
Title:	Organization as a Multi-level Design Pattern for Agent-based Simulation of Complex Systems
Authors:	Vianney Sicard, Mathieu Andraud and Sébastien Picault
Abstract:	This paper describes a generic design pattern to introduce organizational mechanisms into multi-level agent-based simulation architectures, to help the modelling of highly structured complex systems. This pattern makes it possible to specify how to couple any three levels of agents in a multi-level simulation architecture, through their relationships to environments, taking into account organizational constraints. As a proof of concept, we applied this pattern to the fine-grained modelling of batch management in pig farms, and illustrate how the pattern can be instantiated and composed at several agent levels to accurately handle a complex organization in time and space. We thus demonstrate the benefits of combining organizational concepts and multi-level patterns to represent and simulate complex dynamic systems.
Download

Paper Nr:	75
Title:	A City-aware Car Parks Marketplace for Smart Parking
Authors:	Claudia Di Napoli and Silvia Rossi
Abstract:	Searching for a parking space in high populated urban areas is one of the major sources of traffic congestion, increased carbon emission, and wasted time for drivers. In this work, a multi-agent smart parking system is proposed to reserve parking spaces in response to parking requests. It is based on a distributed negotiation mechanism simulating a car park marketplace composed of parking space buyers and sellers. Negotiation is used to obtain parking allocations by taking into account different needs regarding parking location and price, an efficient distribution of parking spaces, and car circulation restrictions. In order to simulate a realistic marketplace, the distributed negotiation mechanism occurs among a set of drivers requesting parking spaces, and a set of parking vendors. The aim of the experimental evaluation is to determine the scalability of a distributed marketplace with respect to parking space re-sellers that share common city policy regulations, to allow for a smart distribution of allocations. The negotiation outcome is experimentally evaluated by considering the resulting social welfare of all the involved negotiators.
Download

Paper Nr:	85
Title:	Network Topology Identification using Supervised Pattern Recognition Neural Networks
Authors:	Aniruddha Perumalla, Ahmet T. Koru and Eric N. Johnson
Abstract:	This paper studies the network topology identification of multi-agent systems with single-integrator dynamics using supervised pattern recognition networks. We split the problem into two classes: (i) small-scale systems, and (ii) large-scale systems. In the small-scale case, we generate all connected (undirected) graphs. A finite family of vectors represent all possible initial conditions by gridding the interval 0 and 1 for each agent. The system responses for all graphs with all initial conditions are the training data for the supervised pattern recognition neural network. This network is successful in identification of the most connected node in up to nearly 99% of cases involving small-scale systems. We present the accuracy of the trained network for network topology identification with respect to grid space. Then, an algorithm predicated on the pattern recognition network, which is trained for a small-scale system, identifies the most connected node in large-scale systems. Monte Carlo simulations estimate the accuracy of the algorithm. We also present the results for these simulations, which demonstrate that the algorithm succeeds in finding the most connected node in more than 60% of the test cases.
Download

Paper Nr:	102
Title:	Strategising RoboCup in Real Time with Uppaal Stratego
Authors:	Philip I. Holler, Magnus K. Jensen, Hannah K. Lockey and Michele Albano
Abstract:	The RoboCup simulator is a playing ground for Agents and Artificial Intelligence research. One of the main challenges provided by RoboCup is generating winning strategies for a set of agents playing soccer, given a partial and noisy view of the game state. Additionally, RoboCup is timing sensitive, and all decisions have to be sent to the server within each tick of 100ms. This paper presents a method for generating strategies by modelling players and scenarios as timed automata in the Uppaal environment. The newest extension of Uppaal, called Uppaal Stratego, allows for synthesising strategies optimising a reward function that is used to guide the decision process. In order to stay within the time frame of 100ms, two approaches were tested, namely forecasting the game state and generating a strategy asynchronously for a later point in time, and generating strategies beforehand and saving them in a lookup table. Four timed automata were developed, and were tested against publicly available methods. We found that strategies could be successfully generated and used within the time constraints of RoboCup using our proposed method. Especially when combined together, our strategies are able to outperform most published methods, but lose against the published world champions.
Download

Paper Nr:	108
Title:	Effective Area Partitioning in a Multi-Agent Patrolling Domain for Better Efficiency
Authors:	Katsuya Hattori and Toshiharu Sugawara
Abstract:	This study proposes a cooperative method for a multi-agent continuous cooperative patrolling problem by partitioning the environment into a number of subareas so that the workload is balanced among multiple agents by allocating subareas to individual agents. Owing to the advancement in robotics and information technology over the years, robots are being utilized in many applications. As environments are usually vast and complicated, a single robot (agent) cannot supervise the entire work. Thus, cooperative work by multiple agents, even though complicated, is indispensable. This study focuses on cooperation in a bottom-up manner by fairly partitioning the environment into subareas, and employing each agent to work on them as its responsibility. However, as the agents do not monitor the entire environment, the decentralized control may generate unreasonable shapes of subareas; the area are often unnecessarily divided into fragmented enclaves, resulting in inefficiency. Our proposed method reduced the number of small and isolated enclaves by negotiation. Our experimental results indicated that our method eliminated the minute/unnecessary fragmented enclaves and improved performance when compared with the results obtained by conventional methods.
Download

Paper Nr:	109
Title:	Decentralized Multi-agent Formation Control via Deep Reinforcement Learning
Authors:	Aniket Gutpa and Raghava Nallanthighal
Abstract:	Multi-agent formation control has been a much-researched topic and while several methods from control theory exist, they require astute expertise to tune properly which is highly resource-intensive and often fails to adapt properly to slight changes in the environment. This paper presents an end-to-end decentralized approach towards multi-agent formation control with the information available from onboard sensors by using a Deep Reinforcement learning framework. The proposed method directly utilizes the raw sensor readings to calculate the agent’s movement velocity using a Deep Neural Network. The approach utilizes Policy gradient methods to generalize efficiently on various simulation scenarios and is trained over a large number of agents. We validate the performance of the learned policy using numerous simulated scenarios and a comprehensive evaluation. Finally, the performance of the learned policy is demonstrated in new scenarios with non-cooperative agents that were not introduced during the training process.
Download

Paper Nr:	124
Title:	Information-theoretic Cost of Decision-making in Joint Action
Authors:	Dari Trendafilov, Daniel Polani and Alois Ferscha
Abstract:	We investigate the information processing cost relative to utility, associated with joint action in dyadic decision-making. Our approach, built on the Relevant Information formalism, combines Shannon’s Information Theory and Markov Decision Processes for modelling dyadic interaction, where two agents with independent controllers move an object together with fully redundant control in a grid world. Results show that increasing collaboration relaxes the pressure on required information intake and vice versa, antagonistic behavior takes a higher toll on information bandwidth. In this trade-off the particular embodiment of the environment plays a key role, demonstrated in simulations with informationally parsimonious optimal controllers.
Download

Paper Nr:	142
Title:	Mixed Reference Interpretation in Multi-turn Conversation
Authors:	Nanase Otake, Shoya Matsumori, Yosuke Fukuchi, Yusuke Takimoto and Michita Imai
Abstract:	Contextual reference refers to the mention of matters or topics that appear in the conversation, and situational reference to the mention of objects or events that exist around the conversation participants. In conventional utterance processing, the system deals with either contextual or situational reference in a dialogue. However, in order to achieve meaningful communication between people and the system in the real world, the system needs to consider Mixed Reference Interpretation (MRI) problem, that is, handling both types of reference in an integrated manner. In this paper, we propose DICONS, a method that sequentially estimates an interpretation of utterances from interpretation candidates derived from both contextual reference and situational reference in a dialogue. In an experiment in which DICONS handled this task with both contextual and situational references, we found that it could properly judge which type of reference had occurred. We also found that the referent of the demonstrative word in each context and situation could be properly estimated.
Download

Paper Nr:	154
Title:	Improving Active Attitude for Interactive Decision-making with Multiple Agents by Increasing Personal Resource
Authors:	Yoshimasa Ohmoto, Masato Kuno and Toyoaki Nishida
Abstract:	When human participants collaborate in decision-making and problem solving, the results are often better than those obtained from individual efforts. However, when they cooperatively perform tasks with an embodied agent, the agent is often regarded merely as a human-centric multi-modal interface from which information is obtained. In this study, by increasing the “personal resource” in the aspect of work engagement, we aim to incorporate active attitude in the human participants toward task and in their interactions with the embodied agents. We conducted an experiment to investigate whether increasing the personal resource will impact the active attitude of participants. In the experiment, we used a mediator agent that increased either the “personal resource” or “job resource” in addition to an expert agent that directly supported the task. The results suggested that a mediator agent that increased the personal resources can induce the active attitude of participants in the human-agent interaction.
Download

Paper Nr:	160
Title:	Towards Verifying a Blocks World for Teams GOAL Agent
Authors:	Alexander B. Jensen
Abstract:	We continue to see an increase in applications based on multi-agent system technology. As the technology becomes more widespread, so does the requirement for agent systems to operate reliably. In this paper, we expand on the approach of using an agents logic to prove properties of agents. Our work describes a transformation from GOAL program code to an agent logic. We apply it to a Blocks World for Teams agent and prove a correctness property. Finally, we sketch future challenges of extending the framework.
Download

Paper Nr:	175
Title:	A Self-adaptive Module for Cross-understanding in Heterogeneous MultiAgent Systems
Authors:	Guilhem Marcillaud, Valérie Camps, Stéphanie Combettes, Marie-Pierre Gleizes and Elsy Kaddoum
Abstract:	We propose a self-adaptive module, called LUDA (Learning Usefulness of DAta) to tackle the problem of cross-understanding in heterogeneous multiagent systems. In this work heterogeneity concerns the agents usage of information available under different reference frames. Our goal is to enable an agent to understand other agents information. To do this, we have built the LUDA module analysing redundant information to improve their accuracy. The closest domains addressing this problem are feature selection and data imputation. Our module is based on the relevant characteristics of these two domains, such as selecting a subset of relevant information and estimating the missing data value. Experiments are conducted using a large variety of synthetic datasets and a smart city real dataset to show the feasibility in a real scenario. The results show an accurate transformation of other information, an improvement of the information use and relevant computation time for agents decision making.
Download

Paper Nr:	181
Title:	Dynamic Lot Sizing in a Self-organizing Production
Authors:	Martin Krockert, Marvin Matthes and Torsten Munkelt
Abstract:	Companies more and more offer individual products to satisfy their customers and stand out from other competitors. Those individual products differ in their production process and thus require many different tool- resource combinations, so called setups. In order to reduce the number of setups, shortening the overall setup time, reducing throughput time, while increasing the adherence to delivery dates, we propose a dynamic lot sizing approach that combines separate operations into so-called buckets. In this paper, we present the implementation of the dynamic lot sizing approach in our multi-agent based self-organizing production, using two different production models, which demonstrate the efficiency of our solution in comprehension to an exhaustive rule to create buckets as a results of an empirical study.
Download

Paper Nr:	184
Title:	Cooperative Neighborhood Learning: Application to Robotic Inverse Model
Authors:	Bruno Dato, Marie-Pierre Gleizes and Frédéric Migeon
Abstract:	In this paper we present a generic multiagent learning system based on context learning applied in robotics. By applying learning with multiagent systems in robotics, we propose an endogenous self-learning strategy to improve learning performances. Inspired by constructivism, this learning mechanism encapsulates models in agents. To enhance the learning performance despite the weak amount of data, local and internal negotiation, also called cooperation, is introduced. Agents collaborate by generating artificial learning situations to improve their model. A second contribution is a new exploitation of the learnt models that allows less training. We consider highly redundant robotic arms to learn their Inverse Kinematic Model. A multiagent system learns a collective of models for a robotic arm. The exploitation of the models allows to control the end position of the robotic arm in a 2D/3D space. We show how the addition of artificial learning situations increases the performances of the learnt model and decreases the required labeled learning data. Experimentations are conducted on simulated arms with up to 30 joints in a 2D task space.
Download

Paper Nr:	200
Title:	A Branch and Price Algorithm for Coalition Structure Generation over Graphs
Authors:	Emanuel F. Olariu, Cristian Frăsinaru and Policiuc A. Albert
Abstract:	This paper presents an integer linear programming approach for the coalition structure generation (CSG) problem over graphs. Forming such structures is a major problem in areas like artificial intelligence an multi-agent systems. The problem asks to partition a given set of agents into coalitions in order to maximize their social well-fare - the agents being vertices in a given graph and their communication links being the edges. We give a truncated branch and price algorithm using valuation functions for which this problem is proven to be computationally hard. We consider three cases: first when the value of a coalition is the sum of the weights of its edges, second when the value takes account of both inter- and intra-coalitional disagreements and agreements, respectively, and another one when the value takes account of the pairs of adjacent agents which have common neighbors outside. The experimental results cover sets of up to fifty agents. Our approach prove that an off the shelf optimization solver can be used to solve CSG problem over graphs for some of the most used valuation functions. We prove also that for the coordination valuation the corresponding decision problem is NP-complete when the number of coalitions must be two.
Download

Paper Nr:	208
Title:	Distributed Serverless Chat Bot Networks using Mobile Agents: A Distributed Data Base Model for Social Networking and Data Analytics
Authors:	Stefan Bosse
Abstract:	Today human-machine dialogues performed and moderated by chat bots are ubiquitous. Commonly, centralised and server-based chat bot software is used to implement rule-based and intelligent dialogue robots. Furthermore, human networking is not supported. Rule-based chat bots typically implement an interface to a knowledge data base in a more natural way. The dialogue topics are narrowed and static. Intelligent chat bots aim to improve dialogues and conversational quality over time and user experience. In this work, mobile agents are used to implement a distributed, decentralised, serverless dialogue robot network that enables ad-hoc communication between humans and machines (networks) and between human groups via the chat bot network (supporting personalized and mass communication). I.e., the chat bot networks aims to extend the communication and social interaction range of humans, especially in mobile environments, by a distributed knowledge and data base approach. Additionally, the chat bot network is a sensor data acquisition and data aggregator system enabling large-scale crowd-based analytics. A first proof-of-concept demonstrator is shown identifying the challenges arising with self-organising distributed chat bot networks in resource-constrained mobile networks. The novelty of this work is a hybrid chat bot multi-agent architecture enabling scalable distributed and adaptive communicating chat bot networks.
Download

Paper Nr:	219
Title:	A Local Active Learning Strategy by Cooperative Multi-Agent Systems
Authors:	Bruno Dato, Marie-Pierre Gleizes and Frédéric Migeon
Abstract:	In this paper, we place ourselves in the context learning approach and we aim to show that adaptive multi-agent systems are a relevant solution to its enhancement with local active learning strategy. We use a local learning approach inspired by constructivism: context learning by adaptive multi-agent systems. We seek to introduce active learning requests as a mean of internally improving the learning process by detecting and resolving imprecisions between the learnt knowledge. We propose a strategy of local active learning for resolving learning inaccuracies. In this article, we evaluate the performance of local active learning. We show that the addition of active learning requests facilitated by self-observation accelerates and generalizes learning, intelligently selects learning data, and increases performance on prediction errors.
Download

Paper Nr:	224
Title:	Heterogeneous Preferences and Patterns of Contribution in Cybersecurity as a Public Good
Authors:	Mazaher Kianpour
Abstract:	This paper presents an agent-based model of contribution to cybersecurity as a participatory public good. Ineffective cybersecurity measures pose serious threats and risks to the development and stability of information societies in the world. Hence, different doctrines and thesis have been suggested to explore how this domain should be treated by the public and private stakeholders. Cybersecurity as a public good is one of these doctrines that accordingly, cybersecurity is non-rivalrous and non-excludable. In this paper, we present a model of social preferences reflecting the concepts of altruism, individualism, aggressiveness, and reciprocity. It describes an agent-based model simulating a repeated public goods game among a set of defenders that are in an uncertain environment with incomplete and imperfect information. In the model, defenders have a probability to choose contribution or being a free-rider, depending on their own preferences and facing with revealed preferences of other defenders. This model implements a utility maximization that applies to each individual, modeling the existence of free-riders, punishments, and interdependency of decisions on the social context. The results of this simulation show that, over time, defenders update their preferences in reaction to the behavior of other defenders and the experience of cyber-attacks. Moreover, they indicate a high level of contribution to the provision of cybersecurity as a public good and the effectiveness of punishment on increasing the contributions. This paper demonstrated how agent-based models can be used to examine this doctrine and investigate whether this doctrine complies with the unique characteristics of cybersecurity.
Download

Paper Nr:	233
Title:	Path Planning for Autonomous Vehicles with Dynamic Lane Mapping and Obstacle Avoidance
Authors:	Ahmed El Mahdawy and Amr El Mougy
Abstract:	Path planning is at the core of autonomous driving capabilities, and obstacle avoidance is a fundamental part of autonomous vehicles as it has a great effect on passenger safety. One of the challenges of path planning is building an accurate map that responds to changes in the drivable area. In this paper, we present a novel path planning model with static and moving obstacle avoidance capabilities, LiDAR-based localization, and dynamic lane mapping according to road width. We describe our cost-based map building approach and show the vehicle trajectory model. Then, we evaluate our model by performing a simulation test as well as a real life demo, in which the proposed model proves to be effective at maneuvering around static road obstacles, as well as avoiding collisions with moving obstacles such as in pedestrian crossing scenarios.
Download

Paper Nr:	236
Title:	Towards a Model of Empathic Pedagogical Agent for Educating Children and Teenagers on Good Practices in the Use of Social Networks
Authors:	Joaquín Taverner, Emilio Vivancos and Vicente Botti
Abstract:	Social networks have been a revolution for our society. The rapid expansion of these networks results in more and more children and teenagers using them regularly. However, despite the fact that most social networks have privacy control systems, most young people do not use these controls because teenagers are usually unaware of the privacy risks that exist in the Internet. This inadequate use of social media can lead to different social problems such as cyber-bullying, grooming, or sexting. The best way to prevent these risks is through education. In this paper we propose an empathic pedagogical agent model for education on good practices in the use of social networks. The agent interacts empathetically with users through the social network. The user’s emotion is recognized through a camera and is processed in real time to obtain the emotion. The agent analyzes the recognized emotion and the users’ actions and looks for the best strategy to advice and educate the teenager in the correct use of the social network.
Download

Paper Nr:	241
Title:	On using Theorem Proving for Cognitive Agent-oriented Programming
Authors:	Alexander B. Jensen, Koen V. Hindriks and Jørgen Villadsen
Abstract:	Demonstrating reliability of cognitive multi-agent systems is of key importance. There has been an extensive amount of work on logics for verifying cognitive agents but it has remained mostly theoretical. Cognitive agent-oriented programming languages provide the tools for compact representation of complex decision making mechanisms, which offers an opportunity for applying a theorem proving approach. We base our work on the belief that theorem proving can add to the currently available approaches for providing assurance for cognitive multi-agent systems. However, a practical approach using theorem proving is missing. We explore the use of proof assistants to make verifying cognitive multi-agent systems more practical.
Download

Paper Nr:	267
Title:	Integrating Shared Information into the Sensorial Mapping of Connected and Autonomous Vehicles
Authors:	Filipo S. Perotto, Stephanie Combettes, Valerie Camps, Elsy Kaddoum, Guilhem Marcillaud, Pierre Glize and Marie-Pierre Gleizes
Abstract:	A connected and autonomous vehicle (CAV) needs to dynamically maintain a map of its environment. Even if the self-positioning and relative localization of static objects (roads, signs, poles, guard-rails, buildings, etc.) can be done with great precision thanks to the help of hd-maps, the detection of the dynamic objects on the scene (other vehicles, bicycles, pedestrians, animals, casual objects, etc.) must be made by the CAV itself based on the interpretation of its low-level sensors (radars, lidars, cameras, etc.). In addition to the need of representing those moving objects around it, the CAV (seen as an agent immersed in that traffic environment) must identify them and understand their behavior in order to anticipate their expected trajectories. The accuracy and completeness of this real-time map, necessary for safely planning its own maneuvers, can be improved by incorporating the information transmitted by other vehicles or entities within the surrounding neighborhood through V2X communications. The implementation of this cooperative perception can be seen as the last phase of perception fusion, after the in-vehicle signals (coming from its diverse sensors) have already been combined. In this position paper, we approach the problem of creating a coherent map of objects by selecting relevant information sent by the neighbor agents. This task requires correctly identifying the position of other communicant agents, based both on the own sensory perception and on the received information, and then correcting and completing the map of perceived objects with the communicated ones. For doing so, the precision and confidence on each information must be taken into account, as well as the trust and latency associated with each source. The broad objective is to model and simulate a fleet of vehicles with different levels of autonomy and cooperation, based on a multi-agent architecture, in order to study and improve road safety, traffic efficiency, and passenger comfort. In the paper, the problem is stated, a brief survey of the state-of-the-art on related topics is given, and the sketch of a solution is proposed.
Download

Paper Nr:	282
Title:	Investigation on Stochastic Local Search for Decentralized Asymmetric Multi-objective Constraint Optimization Considering Worst Case
Authors:	Toshihiro Matsui
Abstract:	The Distributed Constraint Optimization Problem (DCOP) has been studied as a fundamental problem in multiagent cooperation. With the DCOP approach, various cooperation problems including resource allocation and collaboration among agents are represented and solved in a decentralized manner. Asymmetric Multi-Objective DCOP (AMODCOP) is an extended class of DCOPs that formalizes the situations where agents have individual objectives to be simultaneously optimized. In particular, the optimization of the worst case objective value among agents is important in practical problems. Existing works address complete solution methods including extensions with approximation. However, for large-scale and dense problems, such solution methods are insufficient. Although the existing studies also address a few simple deterministic local search methods, there are opportunities to introduce stochastic local search methods. As the basis for applying stochastic local search methods to AMODCOPs for the preferences of agents, we introduce stochastic local search methods with several optimization criteria. We experimentally analyze the influence of the optimization criteria on perturbation in the exploration process of search methods and investigate additional information propagation that extends the knowledge of the agents who are performing the local search.
Download

Paper Nr:	8
Title:	Autonomous Braking and Throttle System: A Deep Reinforcement Learning Approach for Naturalistic Driving
Authors:	Varshit Dubey, Ruhshad Kasad and Karan Agrawal
Abstract:	Autonomous Braking and Throttle control is key in developing safe driving systems for the future. There exists a need for autonomous vehicles to negotiate a multi-agent environment while ensuring safety and comfort. A Deep Reinforcement Learning based autonomous throttle and braking system is presented. For each time step, the proposed system makes a decision to apply the brake or throttle. The throttle and brake are modelled as continuous action space values. We demonstrate 2 scenarios where there is a need for a sophisticated braking and throttle system, i.e when there is a static obstacle in front of our agent like a car, stop sign. The second scenario consists of 2 vehicles approaching an intersection. The policies for brake and throttle control are learned through computer simulation using Deep deterministic policy gradients. The experiment shows that the system not only avoids a collision, but also it ensures that there is smooth change in the values of throttle/brake as it gets out of the emergency situation and abides by the speed regulations, i.e the system resembles human driving.
Download

Paper Nr:	25
Title:	Smart Broker Agent Learning How to Reach Appropriate Goal by Making Appropriate Compromises
Authors:	Dilyana Budakova, Veselka Petrova-Dimitrova and Lyudmil Dakovski
Abstract:	In this paper a new Smart Broker Learning Agent (SBLA) has been proposed, which trains to find the most acceptable solution to a given problem, according to the individual requirements and emotions of a particular user. For this purpose, a new structure of the agent has been proposed and reinforcement-learning algorithm has been used. When the scenarios and criteria under consideration are complex, and when mixed emotions arise, it may be necessary to compromise on certain criteria in order to achieve the goal. Then knowledge of the preferences and emotions of the particular user is needed. In these cases, the SBLA does not allow compromises that are unacceptable to this user. The structure and the way of acting of the agent have been considered. The knowledge that the SBLA must have and the process of its formation have been described. The scenarios for solving a specific task and the conducted experiments have been presented. Some contributions, arising from the use of the proposed agent’s architecture have been discussed, such as: the opportunity for the agent to explain decisions; to offer the most appropriate solution for each specific user; to avoid unacceptable compromises, to have empathy, and the greater approval of the offered solutions.
Download

Paper Nr:	30
Title:	Simulation Framework to Train Intelligent Agents towards an Assisted Driving Power Wheelchair for People with Disability
Authors:	Giovanni Falzone, Gianluca Giuffrida, Silvia Panicacci, Massimiliano Donati and Luca Fanucci
Abstract:	Several million people with disabilities exploit power wheelchairs for outdoor mobility on both sidewalks and cycling paths. Especially those with upper limb motor impairments have difficulty reacting quickly to obstacles along the way, creating dangerous situations, such as wheelchair crash or rollover. A possible solution could be to equip the power wheelchair with a neural network-based assisted driving system, able to detect, avoid or warn the users of obstacles. Therefore, a virtual environment is required to simulate the system and then test different neural network architectures before mounting the best performing one directly on board. In this work, we present a simulation framework to train multiple artificial intelligent agents in parallel, by means of reinforcement learning algorithms. The agent shall follow the user’s will and identify obstacles along the path, taking the control of the power wheelchair when the user is making a dangerous driving choice. The developed framework, adapted from an existing autonomous driving simulator, has been used to train and test multiple intelligent agents simultaneously, thanks to a customised synchronisation and memory management mechanism, reducing the overall training time. Preliminary results highlight the suitability of the adapted framework for multiple agent development in the assisted driving scenario.
Download

Paper Nr:	40
Title:	Imperfect Oracles: The Effect of Strategic Information on Stock Markets
Authors:	Miklos Borsi
Abstract:	Modern financial market dynamics warrant detailed analysis due to their significant impact on the world. This, however, often proves intractable; massive numbers of agents, strategies and their change over time in reaction to each other leads to difficulties in both theoretical and simulational approaches. Notable work has been done on strategy dominance in stock markets with respect to the ratios of agents with certain strategies. Perfect knowledge of the strategies employed could then put an individual agent at a consistent trading advantage. This research reports the effects of imperfect oracles on the system - dispensing noisy information about strategies - information which would normally be hidden from market participants. The effect and achievable profits of a singular trader with access to an oracle were tested exhaustively with previously unexplored factors such as changing order schedules. Additionally, the effect of noise on strategic information was traced through its effect on trader efficiency.
Download

Paper Nr:	53
Title:	The Gopher’s Gambit: Survival Advantages of Artifact-based Intention Perception
Authors:	Cynthia Hom, Amani R. Maina-Kilaas, Kevin Ginta, Cindy Lay and George D. Montañez
Abstract:	Being able to assess and calculate risks can positively impact an agent’s chances of survival. When other intelligent agents alter environments to create traps, the ability to detect such intended traps (and avoid them) could be life-saving. We investigate whether there are cases for which an agent’s ability to perceive intention through the assessment of environmental artifacts provides a measurable survival advantage. Our agents are virtual gophers assessing a series of room-like environments, which are potentially dangerous traps intended to harm them. Using statistical hypothesis tests based on configuration coherence, the gophers differentiate between designed traps and configurations that are randomly generated and most likely safe, allowing them access to the food contained within them. We find that gophers possessing the ability to perceive intention have significantly better survival outcomes than those without intention perception in most of the cases evaluated.
Download

Paper Nr:	55
Title:	Evaluating Correlations in IoT Sensors for Smart Buildings
Authors:	Davide A. Guastella, Nicolas Verstaevel, Cesare Valenti, Bilal Arshad and Johan Barthélemy
Abstract:	In this paper we introduce a dataset of environmental information obtained via indoor and outdoor sensors deployed in the SMART Infrastructure Facility of the University of Wollongong (Australia). The acquired dataset is also made open-sourced along with this paper. We also propose a novel approach based on an evolutionary algorithm to determine pairs of correlated sensors. We compare our approach with three other standard techniques on the same dataset: on average, the accuracy of the evolutionary method is about 62,92%. We also evaluate the computational time, assessing the suitability of the proposed pipeline for real-time applications.
Download

Paper Nr:	77
Title:	Blockchain-based Task-centric Team Building
Authors:	Alexander Jahl, Stefan Jakob, Harun Baraki, Yasin Alhamwy and Kurt Geihs
Abstract:	Large-scale dynamic environments like Industry 4.0, Smart Cities, and Search & Rescue missions require a distributed and effective management of participating autonomous units. Usually, these units and their capabilities are heterogeneous and partially unknown at design time. Thus, the management has to adapt dynamically to the current situation. Several units have to collaborate to solve common tasks, and thus have to share their knowledge. However, complex tasks typically require the splitting of a team of units into subteams that solve smaller subtasks. A common approach to tackle this problem is to employ a decentralised, self-organising system. Traditionally, such systems are modelled either agent-centric or organisation-centric. In contrast, in this paper we shift the focus to a task-centric view. Tasks are enabled to search and bind suitable execution units based on their capabilities. These units can be either single agents, teams of agents, or teams of teams. A blockchain-based allocation model supports the task-centric view and controls the distributed task assignment. We present a proof-of-concept implementation that shows the viability of our presented approach.
Download

Paper Nr:	98
Title:	Comparative Evaluation of Road Traffic Simulators based on Modeler’s Specifications: An Application to Intermodal Mobility Behaviors
Authors:	Azise O. Diallo, Guillaume Lozenguez, Arnaud Doniec and René Mandiau
Abstract:	Today, large cities and peri-urban areas experience problems in the mobility of their population. Faced with this problem, decision-makers must have reliable tools to help them to build and evaluate their policies of mobility. Computer simulations especially traffic simulation tools are, therefore, the solution to better understand (study) the problem and test different resolution scenarios. Unfortunately, there are numerous simulation tools and the choice can be very difficult for traffic modelers. In this paper, we present, based on a generic method, a comparison of the most popular traffic simulation tools in two steps: 1) a comparison part using a weighted system of evaluation criteria to automatically select the candidate simulators. 2) a deeper study of the candidate simulators according to a simulation scenario corresponding to the study case. Finally, this paper presents an application of this method for the selection of a simulator for the study of intermodal mobility behaviors where MATSIM and SUMO were studied in deeper.
Download

Paper Nr:	112
Title:	An Artificial Hormone-based Algorithm for Production Scheduling from the Bottom-up
Authors:	Wilfried Elmenreich, Alexander Schnabl and Melanie Schranz
Abstract:	This paper presents a model for supporting a production scheduling system with an artificial hormone algorithm. The system consists of lots that have to undergo a number of processing steps on different machines. The processing steps for a lot are formalized in a recipe assigned to the lot type. Since the steps in the recipe have to be processed in order, the given system allows choice only in the context of selecting a particular machine for the next step and in changing the processing order of waiting lots at a machine. Optimization of such a job-shop scheduling system is an NP-hard problem. In the approach proposed by this paper, artificial hormone systems are used to express the urgency of a lot and the need for new lots at a machine type, thus providing a system using local information for optimization. Results indicate that the artificial hormone system provides an improvement of around 5% over a First Come-First Serve approach.
Download

Paper Nr:	134
Title:	Bio-Inspired Protocols for Embodied Multi-Agent Systems
Authors:	Vinicius Souza de Jesus, Carlos E. Pantoja, Fabian Manoel, Gleifer V. Alves, Jose Viterbo and Eduardo Bezerra
Abstract:	Bio-Inspired approaches and techniques are being used in different domains and applications in artificial intelligence, including the agent domain. Some agents are able to move from one system to another to establish new relationships. In biology, ecological relations are concepts responsible for classifying the relationships between living beings in an ecosystem, depending on the behavior and function that each one can assume. The objective of this work is to propose bio-inspired protocols based on ecological relations: Predation, Inquilinism, and Mutualism. These protocols aims to preserve agents’ knowledge as they can live as a tenant in another physical body waiting for a similar hardware to predate, or acquire and transmit knowledge by interacting with other agents while sharing the same physical body. To validate these protocols, a study case and a scenario are implemented, tested, and evaluated in a real environment.
Download

Paper Nr:	161
Title:	Towards Verifying GOAL Agents in Isabelle/HOL
Authors:	Alexander B. Jensen
Abstract:	The need to ensure reliability of agent systems increases with the applications of multi-agent technology. As we continue to develop tools that make verification more accessible to industrial applications, it becomes an even more critical requirement for the tools themselves to be reliable. We suggest that this reliability ought not be based on empirical evidence such as testing procedures. Instead we propose using an interactive theorem prover to ensure the reliability of the verification process. Our work aims to verify agent systems by emdedding a verification framework in the interactive theorem prover Isabelle/HOL.
Download

Paper Nr:	193
Title:	A Lightweight Microservice-oriented Platform for Development of Intelligent Agent-based Enterprise Applications
Authors:	Aluizio Haendchen Filho, Rafael C. Ribeiro, Hércules Antônio do Prado, Edilson Ferneda and Jeferson M. Thalheimer
Abstract:	Excess of adherence to the agent technology standards can hinder the software development process while the focus in good practices can lever this process. This paper presents IDEA, a lightweight microservice-oriented platform that facilitates the development and execution of Multi-Agent Systems in the business context. The solution design seeks for a good trade-off between usability and adherence to the agent technology standards. The platform also enables microservices discovering, composition, and reuse.
Download

Paper Nr:	195
Title:	Maintaining Organizational Multi-agent Systems: A Reorganization-based Preventive Approach
Authors:	Nawel Ghrieb, Farid Mokhati and Tahar Guerram
Abstract:	In this article, we propose a preventive maintenance approach for Organization-Centered Multi-Agent System (OCMAS). This approach is based on the quality assessment for the maintenance of OCMAS. The quality of OCMAS is monitored by Aspect Oriented Programming (AOP) techniques, in order to detect any abnormal regression in the quality of the system or that of the agents composing it. This degradation in quality is, usually, an indication of problems that may arise in the structure of the organization or its functionalities. In the context of this work, we are interested in the functional problems that can affect the system and we treated them by reallocating agents to roles. This reallocation is, generally, necessary when the agent is not able to achieve its objectives, which leads to a degradation of the overall quality of the system. Our maintenance approach must therefore anticipate these problems and react by reorganizing the running system in order to improve its quality and allow it to resume its normal behaviour.
Download

Paper Nr:	232
Title:	Distributed Framework for Reversible Merging of Heterogeneous Robot Maps
Authors:	Ilze Andersone
Abstract:	Studies have shown that multi-robot mapping has the benefit of faster environment exploration when compared to single robot mapping. However, when multiple robots explore the environment simultaneously, a new problem arises – how to merge the individual robot maps. While there are many map merging methods developed for homogeneous maps, heterogeneous robot map merging is still a new research area. Another relatively little researched aspect of map merging is how to deal with an error in the map merging decision. This paper proposes a map merging framework for the distributed merging of heterogeneous robot maps and offers two approaches for the further mapping with an emphasis on map merging process reversibility.
Download