ICAART 2024 Abstracts

Area 1 - Artificial Intelligence
Full Papers Short Papers
Area 2 - Agents
Full Papers Short Papers

Area 1 - Artificial Intelligence

Full Papers
Paper Nr: 12
Title:

Deep Reinforcement Learning and Transfer Learning Methods Used in Autonomous Financial Trading Agents
Authors:

Ciprian Paduraru, Catalina C. Patilea and Stefan Iordache
Abstract: It is reported that some of the largest companies from the banking and business sectors are investing massively in the field of trading with automated methods. The methods used vary from classical time series based methods to Deep Learning and more recently Reinforcement Learning (RL). The main goal of this work is first to improve the state of the art in RL-based trading agents. Then, we focus on evaluating the robustness of the trained agents when they are transferred to different trading markets than the ones they were trained on. The framework we developed, RL4FIN, is open source and can be tested by both academia and industry. The evaluation section shows the improvements over state-of-the-art using some public datasets.
Download
Paper Nr: 17
Title:

Adaptive Action Supervision in Reinforcement Learning from Real-World Multi-Agent Demonstrations
Authors:

Keisuke Fujii, Kazushi Tsutsui, Atom Scott, Hiroshi Nakahara, Naoya Takeishi and Yoshinobu Kawahara
Abstract: Modeling of real-world biological multi-agents is a fundamental problem in various scientific and engineering fields. Reinforcement learning (RL) is a powerful framework to generate flexible and diverse behaviors in cyberspace; however, when modeling real-world biological multi-agents, there is a domain gap between behaviors in the source (i.e., real-world data) and the target (i.e., cyberspace for RL), and the source environment parameters are usually unknown. In this paper, we propose a method for adaptive action supervision in RL from real-world demonstrations in multi-agent scenarios. We adopt an approach that combines RL and supervised learning by selecting actions of demonstrations in RL based on the minimum distance of dynamic time warping for utilizing the information of the unknown source dynamics. This approach can be easily applied to many existing neural network architectures and provide us with an RL model balanced between reproducibility as imitation and generalization ability to obtain rewards in cyberspace. In the experiments, using chase-and-escape and football tasks with the different dynamics between the unknown source and target environments, we show that our approach achieved a balance between the reproducibility and the generalization ability compared with the baselines. In particular, we used the tracking data of professional football players as expert demonstrations in football and show successful performances despite the larger gap between behaviors in the source and target environments than the chase-and-escape task.
Download
Paper Nr: 25
Title:

Hybrid Quanvolutional Echo State Network for Time Series Prediction
Authors:

Rebh Soltani, Emna Benmohamed and Hela Ltifi
Abstract: Quantum Machine Learning (QML) combines quantum physics with machine learning techniques to enhance algorithm performance. By leveraging the unique properties of quantum computing, such as superposition and entanglement, QML aims to solve complex problems beyond the capabilities of classical computing. In this study, we developed a hybrid model, the quantum convolutional Echo State Network, which incorporates QML principles into the Reservoir Computing framework. Evaluating its performance on benchmark time-series datasets, we observed improved results in terms of mean square error (MSE) and reduced time complexity compared to the classical Echo State Network (ESN). These findings highlight the potential of QML to advance time-series prediction and underscore the benefits of merging quantum and machine learning approaches.
Download
Paper Nr: 31
Title:

VP-DARTS: Validated Pruning Differentiable Architecture Search
Authors:

Tai-Che Feng and Sheng-De Wang
Abstract: Recently Differentiable Architecture Search (DARTS) has gained increasing attention due to its simplicity and efficient search capability. However, such search methods have a significant chance of encountering overfitting, which can result in the performance collapse problem of the discovered models. In this paper, we proposed VP-DARTS, a validated pruning-based differentiable architecture search method using soft pruning, to address this issue. Firstly, unlike previous search methods, we consider the differentiable architecture search process as a model pruning problem. It prunes or removes unimportant operations from the supernet that contains all possible architectures to obtain the final model. We also show that the traditional hard pruning method would gradually reduce the capacity of the search space during training, leading to local optimal results. To get better architectures than hard pruning, we proposed using a parameterized soft pruning approach in our training process. Secondly, the original DARTS method selects the operation with the maximum architecture parameter on each edge to form the final architecture after training. But we found that this approach cannot truly reflect their importance. Therefore, we estimate the impact on the supernet of each candidate operation by using a subset of the validation set to evaluate its degree of importance. Finally, we implement our method on the NAS-Bench-201 search space, and the experimental results show that VP-DARTS is a robust search method that can obtain architectures with good performance and stable results.
Download
Paper Nr: 35
Title:

Lazy Nested Monte Carlo Search for Coalition Structure Generation
Authors:

Milo Roucairol, Jérôme Arjonilla, Abdallah Saffidine and Tristan Cazenave
Abstract: This paper explores Monte-Carlo Search algorithms applied to Multiagent Systems (MAS), specifically focusing on the problem of Coalition Structure Generation (CSG). CSG is a NP-Hard problem consisting in partitioning agents into coalitions to optimize collective performance. Our study makes three contributions: (i) a novel action space representation tailored for CSG, (ii) a comprehensive comparative analysis of multiple algorithms, and the introduction of Lazy NMCS, (iii) a cutting-edge method that surpasses previous benchmarks. By outlining efficient coalition formation strategies, our findings offer insights for advancing MAS research and practical applications.
Download
Paper Nr: 39
Title:

Identification of Opinion and Ground in Customer Review Using Heterogeneous Datasets
Authors:

Po-Min Chuang, Kiyoaki Shirai and Natthawut Kertkeidkachorn
Abstract: Online reviews are a valuable source of information for both potential buyers and enterprises, but not all reviews provide us helpful information. This paper aims at the identification of a user’s opinion and its reason or ground in a review, supposing that a review including a ground for an opinion is helpful. A classifier to identify an opinion and a ground, called the opinion-ground classifier, is trained from three heterogeneous datasets. The first is the existing dataset for discourse analysis, KWDLC, which is the manually labeled but out-domain dataset. The second is the in-domain but weakly supervised dataset made by a rule-based method that checks the existence of causality discourse markers. The third is another in-domain dataset augmented by ChatGPT, where a prompt to generate new samples is given to ChatGPT. We train several models as the opinion-ground classifier. Results of our experiments show that the use of automatically constructed datasets significantly improves the classification performance. The F1-score of our best model is 0 .71, which is 0.12 points higher than the model trained from the existing dataset only.
Download
Paper Nr: 41
Title:

A Challenging Data Set for Evaluating Part-of-Speech Taggers
Authors:

Mattias Wahde, Minerva Suvanto and Marco D. Vedova
Abstract: We introduce a novel, challenging test set for part-of-speech (POS) tagging, consisting of sentences in which only one word is POS-tagged. First derived from Wiktionary, and then manually curated, it is intended as an out-of-sample test set for POS taggers trained over larger data sets. Sentences were selected such that at least one of four standard benchmark taggers would incorrectly tag the word under consideration for a given sentence, thus identifying challenging instances of POS tagging. Somewhat surprisingly, we find that the benchmark taggers often fail on rather straightforward instances of POS tagging, and we analyze these failures in some detail. We also compute the performance of a state-of-the-art DNN-based POS tagger over our set, obtaining an accuracy of around 0.87 for this out-of-sample test, far below its reported performance in the literature. Also for this tagger, we find instances of failure even in rather simple cases.
Download
Paper Nr: 49
Title:

Dataset Characteristics and Their Impact on Offline Policy Learning of Contextual Multi-Armed Bandits
Authors:

Piotr Januszewski, Dominik Grzegorzek and Paweł Czarnul
Abstract: The Contextual Multi-Armed Bandits (CMAB) framework is pivotal for learning to make decisions. However, due to challenges in deploying online algorithms, there is a shift towards offline policy learning, which relies on pre-existing datasets. This study examines the relationship between the quality of these datasets and the performance of offline policy learning algorithms, specifically, Neural Greedy and NeuraLCB. Our results demonstrate that NeuraLCB can learn from various datasets, while Neural Greedy necessitates extensive coverage of the action-space for effective learning. Moreover, the way data is collected significantly affects offline methods’ efficiency. This underscores the critical role of dataset quality in offline policy learning.
Download
Paper Nr: 54
Title:

Improving Parameter Training for VQEs by Sequential Hamiltonian Assembly
Authors:

Jonas Stein, Navid Roshani, Maximilian Zorn, Philipp Altmann, Michael Kölle and Claudia Linnhoff-Popien
Abstract: A central challenge in quantum machine learning is the design and training of parameterized quantum circuits (PQCs). Similar to deep learning, vanishing gradients pose immense problems in the trainability of PQCs, which have been shown to arise from a multitude of sources. One such cause are non-local loss functions, that demand the measurement of a large subset of involved qubits. To facilitate the parameter training for quantum applications using global loss functions, we propose a Sequential Hamiltonian Assembly (SHA) approach, which iteratively approximates the loss function using local components. Aiming for a prove of principle, we evaluate our approach using Graph Coloring problem with a Varational Quantum Eigensolver (VQE). Simulation results show, that our approach outperforms conventional parameter training by 29.99% and the empirical state of the art, Layerwise Learning, by 5.12% in the mean accuracy. This paves the way towards locality-aware learning techniques, allowing to evade vanishing gradients for a large class of practically relevant problems.
Download
Paper Nr: 59
Title:

CNNs Sparsification and Expansion for Continual Learning
Authors:

Basile Tousside, Jörg Frochte and Tobias Meisen
Abstract: Learning multiple sequentially arriving tasks without forgetting previous knowledge, known as Continual Learning (CL), remains a long-standing challenge for neural networks. Most existing CL methods rely on data replay. However, they are not applicable when past data is unavailable or is not allowed to be synthetically generated. To address this challenge, we propose Sparification and Expansion-based Continual Learning (SECL). SECL avoids forgetting of previous tasks by ensuring the stability of the CNN via a stability regularization term, which prevents filters detected as important for past tasks to deviate too much when learning a new task. On top of that, SECL makes the network plastic via a plasticity regularization term that leverage the over-parameterization of CNNs to efficiently sparsify the network and tunes unimportant filters making them relevant for future tasks. Also, SECL enhances the plasticity of the network through a simple but effective heuristic mechanism that automatically decides when and where (at which layers) to expand the network. Experiments on popular CL vision benchmarks show that SECL leads to significant improvements over state-of-the-art method in terms of overall CL performance, as measured by classification accuracy as well as in terms of avoiding catastrophic forgetting.
Download
Paper Nr: 66
Title:

Metrics for Popularity Bias in Dynamic Recommender Systems
Authors:

Valentijn Braun, Debarati Bhaumik and Diptish Dey
Abstract: Albeit the widespread application of recommender systems (RecSys) in our daily lives, rather limited research has been done on quantifying unfairness and biases present in such systems. Prior work largely focuses on determining whether a RecSys is discriminating or not but does not compute the amount of bias present in these systems. Biased recommendations may lead to decisions that can potentially have adverse effects on individuals, sensitive user groups, and society. Hence, it is important to quantify these biases for fair and safe commercial applications of these systems. This paper focuses on quantifying popularity bias that stems directly from the output of RecSys models, leading to over recommendation of popular items that are likely to be misaligned with user preferences. Four metrics to quantify popularity bias in RescSys over time in dynamic setting across different sensitive user groups have been proposed. These metrics have been demonstrated for four collaborative filtering based RecSys algorithms trained on two commonly used benchmark datasets in the literature. Results obtained show that the metrics proposed provide a comprehensive understanding of growing disparities in treatment between sensitive groups over time when used conjointly.
Download
Paper Nr: 68
Title:

Investigation into the Training Dynamics of Learned Optimizers
Authors:

Jan Sobotka, Petr Šimánek and Daniel Vašata
Abstract: Optimization is an integral part of modern deep learning. Recently, the concept of learned optimizers has emerged as a way to accelerate this optimization process by replacing traditional, hand-crafted algorithms with meta-learned functions. Despite the initial promising results of these methods, issues with stability and generalization still remain, limiting their practical use. Moreover, their inner workings and behavior under different conditions are not yet fully understood, making it difficult to come up with improvements. For this reason, our work examines their optimization trajectories from the perspective of network architecture symmetries and parameter update distributions. Furthermore, by contrasting the learned optimizers with their manually designed counterparts, we identify several key insights that demonstrate how each approach can benefit from the strengths of the other.
Download
Paper Nr: 77
Title:

HierNet: Image Recognition with Hierarchical Convolutional Networks
Authors:

Levente Tempfli and Csanád Sándor
Abstract: Convolutional Neural Networks (CNNs) have proven to be an effective method for image recognition due to their ability to extract features and learn the internal representation of the input data. However, traditional CNNs disregard the hierarchy of the input data, which can lead to suboptimal performance. In this paper, we propose a novel method of organizing a CNN into a quasi-decision tree, where the edges represent the feature-extracting layers of a CNN and the nodes represent the classifiers. The structure of the decision tree corresponds to the hierarchical relationships between the label classes, meaning that the visually similar classes are located in the same subtree. We also introduce a simple semi-supervised method to determine these hierarchical relations to avoid having to manually construct such a hierarchy between a large number of classes. We evaluate our method on the CIFAR-100 dataset using ResNet as our base CNN model. Our results show that the proposed method outperforms this base CNN between 2.12-3.77% (depending on the version of the architecture), demonstrating the effectiveness of incorporating input hierarchy into CNNs. Code is available at https://github.com/levtempfli/HierNet.
Download
Paper Nr: 78
Title:

Garment Returns Prediction for AI-Based Processing and Waste Reduction in E-Commerce
Authors:

Marie Niederlaender, Aena N. Lodi, Soeren Gry, Rajarshi Biswas and Dirk Werth
Abstract: Product returns are an increasing burden for manufacturers and online retailers across the globe, both economically and ecologically. Especially in the textile and fashion industry, on average more than half of the ordered products are being returned. The first step towards reducing returns and being able to process unavoidable returns effectively, is the reliable prediction of upcoming returns at the time of order, allowing to estimate inventory risk and to plan the next steps to be taken to resell and avoid destruction of the garments. This study explores the potential of 5 different Machine Learning Algorithms combined with regualised target encoding for categorical features to predict returns of a German online retailer, exclusively selling festive dresses and garments for special occasions, where a balanced accuracy of up to 0.86 can be reached even for newly introduced products, if historical data on customer behavior is available. This work aims to be extended towards an AI-based recommendation system to find the ecologically and economically best processing strategy for garment returns to reduce waste and the financial burden on retailers.
Download
Paper Nr: 79
Title:

GNNDLD: Graph Neural Network with Directional Label Distribution
Authors:

Chandramani Chaudhary, Nirmal K. Boran, N. Sangeeth and Virendra Singh
Abstract: By leveraging graph structure, Graph Neural Networks (GNN) have emerged as a useful model for graph-based datasets. While it is widely assumed that GNNs outperform basic neural networks, recent research shows that for some datasets, neural networks outperform GNNs. Heterophily is one of the primary causes of GNN performance degradation, and many models have been proposed to handle it. Furthermore, some intrinsic information in graph structure is often overlooked, such as edge direction. In this work, we propose GNNDLD, a model which exploits the edge direction and label distribution around a node in varying neighborhoods (hop-wise). We combine features from all layers to retain both low-pass frequency and high-pass frequency components of a node because different layers of neural networks provide different types of information. In addition, to avoid oversmoothing, we decouple the node feature aggregation and transformation operations. By combining all of these concepts, we present a simple yet very efficient model. Experiments on six standard real-world datasets show the superiority of GNNDLD over the state-of-the-art models in both homophily and heterophily.
Download
Paper Nr: 83
Title:

Exploring Unsupervised Anomaly Detection with Quantum Boltzmann Machines in Fraud Detection
Authors:

Jonas Stein, Daniëlle Schuman, Magdalena Benkard, Thomas Holger, Wanja Sajko, Michael Kölle, Jonas Nüßlein, Leo Sünkel, Olivier Salomon and Claudia Linnhoff-Popien
Abstract: Anomaly detection in Endpoint Detection and Response (EDR) is a critical task in cybersecurity programs of large companies. With rapidly growing amounts of data and the omnipresence of zero-day attacks, manual and rule-based detection techniques are no longer eligible in practice. While classical machine learning approaches to this problem exist, they frequently show unsatisfactory performance in differentiating malicious from benign anomalies. A promising approach to attain superior generalization compard to currently employed machine learning techniques is using quantum generative models. Allowing for the largest representation of data on available quantum hardware, we investigate Quantum-Annealing-based Quantum Boltzmann Machines (QBMs) for the given problem. We contribute the first fully unsupervised approach for the problem of anomaly detection using QBMs and evaluate its performance on an EDR-inspired synthetic dataset. Our results indicate that QBMs can outperform their classical analog (i.e., Restricted Boltzmann Machines) in terms of result quality and training steps in special cases. When employing Quantum Annealers from D-Wave Systems, we conclude that either more accurate classical simulators or substantially more QPU time is needed to conduct the necessary hyperparameter optimization allowing to replicate our simulation results on quantum hardware.
Download
Paper Nr: 104
Title:

Multi-Granular Evaluation of Diverse Counterfactual Explanations
Authors:

Yining Yuan, Kevin McAreavey, Shujun Li and Weiru Liu
Abstract: As a popular approach in Explainable AI (XAI), an increasing number of counterfactual explanation algorithms have been proposed in the context of making machine learning classifiers more trustworthy and transparent. This paper reports our evaluations of algorithms that can output diverse counterfactuals for one instance. We first evaluate the performance of DiCE-Random, DiCE-KDTree, DiCE-Genetic and Alibi-CFRL, taking XGBoost as the machine learning model for binary classification problems. Then, we compare their suggested feature changes with feature importance by SHAP. Moreover, our study highlights that synthetic counterfactuals, drawn from the input domain but not necessarily the training data, outperform native counter-factuals from the training data regarding data privacy and validity. This research aims to guide practitioners in choosing the most suitable algorithm for generating diverse counterfactual explanations.
Download
Paper Nr: 109
Title:

Multiple Agents Dispatch via Batch Synchronous Actor Critic in Autonomous Mobility on Demand Systems
Authors:

Jiyao Li and Vicki H. Allan
Abstract: Autonomous Mobility on Demand (AMoD) systems are a promising area in the emerging field of intelligent transportation systems. In this paper, we focus on the problem of how to dispatch a fleet of autonomous vehicles (AVs) within a city while balancing supply and demand. We first formulate the problem as a Markov Decision Process (MDP) of which the goal is to maximize the accumulated average reward, then propose the Multiagent Reinforcement Learning (MARL) framework. The Temporal-Spatial Dispatching Network (TSD-Net) that combines both policy and value network learns representation features facilitating spatial information with its temporal signals. The Batch Synchronous Actor Critic (BS-AC) samples experiences from the Rollout Buffer with replacement and trains parameters of the TSD-Net. Based on the state value from the TSD-Net, the Priority Destination Sampling Assignment (PDSA) algorithm defines orders’ priority by their destinations. Popular destinations are preferred as it is easier for agents to find future work in a popular location. Finally, with the real-world city scale dataset from Chicago, we compare our approach to several competing baselines. The results show that our method is able to outperform other baseline methods with respect to effectiveness, scalability, and robustness.
Download
Paper Nr: 115
Title:

Knowledge Modelling, Strategy Designing, and Agent Engineering for Reconnaissance Blind Chess
Authors:

Robin Stöhr, Shuai Wang and Zhisheng Huang
Abstract: Reconnaissance Blind Chess (RBC) is a unique chess variant where players have limited visibility of a 3x3 square in each round. This paper offers a comparative analysis of the performance of extant agents, along with an assessment of their ability to model their opponents’ knowledge. On the basis of our analytical findings, we propose novel and efficient sensing and movement strategies. Subsequently, these strategies are tested through agent-based gameplay. Furthermore, our experimentation extends to the inference of new knowledge through a strategy based on the Theory of Mind. Collectively, these insights contribute to the selection of the most promising strategies for the design of our Scorca agent. By the time of the paper’s submission, it occupies the second position on the global leaderboard for the RBC game. To conclude, we engage in a discussion of the inherent limitations of the extant agents and offer a glimpse into potential future strategies.
Download
Paper Nr: 120
Title:

Fingerprint Large Classification Using Sequential Learning on Parallel Environment
Authors:

Nicolás A. Reyes-Reyes, Marcela C. González-Araya and Wladimir E. Soto-Silva
Abstract: Fingerprint classification allows a biometric identification system to reduce search space in databases and therefore response times. In the literature, fingerprint classification has been addressed through different approaches where deep learning techniques such as convolutional neural networks have been gaining attention. However, the proposed approaches use extremely small data sets for large-scale real-world scenarios that could worsen accuracy rates due to interclass and intraclass variations in fingerprints. For this reason, we proposed a fingerprint classification approach that allows us to address this problem by considering millions of samples. For this purpose, a classifier based on neural networks trained using online sequential extreme learning machines was developed. Likewise, to accelerate the training of the classifier, the matrix operations inside it was run in a graphic processing unit. In order to evaluate our proposal, the approach was tested on three datasets with more than two million synthetic fingerprint image descriptors. The results are similar in terms of accuracy and computational time to recent approaches but using more than 2.5 million samples.
Download
Paper Nr: 122
Title:

DRL4HFC: Deep Reinforcement Learning for Container-Based Scheduling in Hybrid Fog/Cloud System
Authors:

Ameni Kallel, Molka Rekik and Mahdi Khemakhem
Abstract: The IoT-based applications have a set of complex requirements, such as a reliable network connection and handling data from multiple sources quickly and accurately. Therefore, combining a Fog environment with a Cloud environment can be beneficial for IoT-based applications, as it provides a distributed computing system that can handle large amounts of data in real time. However, the microservice provision to execute such applications with achieving a high Quality of Service (QoS) and low bandwidth communications. Thus, the container-based microservice scheduling problem in a hybrid Fog and Cloud environment is a complex issue that has yet to be fully solved. In this work, we first propose a container-based microservice scheduling model for a hybrid architecture. Our model is a multi-objective scheduler, named DRL4HFC, for Hybrid Fog/Cloud architecture. It is based on two Deep Reinforce Learning (DRL) agents. DRL-based agents learn the inherent properties of the various microservices, nodes, and environments to determine the appropriate placement of each microservice instance required to execute each task within the Business Process (BP). Our proposal aims to reduce the execution time, compute and network resource consumption, and resource occupancy rates of Fog/Cloud nodes. Second, we present a set of experiments in order to evaluate the effectiveness of our algorithm in terms of cost, quality, and time. The experimental results demonstrate that DRL4HFC achieves faster execution times, lower communication costs and better balanced resource loads.
Download
Paper Nr: 127
Title:

An Analysis of Knowledge Representation for Anime Recommendation Using Graph Neural Networks
Authors:

Yuki Saito, Shusaku Egami, Yuichi Sei, Yasuyuki Tahara and Akihiko Ohsuga
Abstract: In recent years, entertainment content, such as movies, music, and anime, has been gaining attention due to the stay-at-home demand caused by the expansion of COVID-19. In the content domain, research in the field of knowledge representation is primarily concerned with accurately describing metadata. Therefore, different knowledge representations are required for applications in downstream tasks. In this study, we aim to clarify effective knowledge representation through a case study of recommending anime works. Thus, we hypothesized how to represent anime works knowledge to improve recommendation performance from both quantitative and qualitative aspects and verified the hypotheses by changing the knowledge representation structure according to the hypothesis. Initially, we collected data about anime works from multiple data sources and integrated them to construct a knowledge graph (KG). We also prepared several KGs by varying the knowledge configuration. Subsequently, we compared the recommendation performance of each KG as an input to the graph neural networks. As a result, it was found that the amount of semantic relationships was proportional to the recommendation performance and that the properties that can characterize the work contributed to the recommendation.
Download
Paper Nr: 129
Title:

Foundations of Dispatchability for Simple Temporal Networks with Uncertainty
Authors:

Luke Hunsberger and Roberto Posenato
Abstract: Simple Temporal Networks (STNs) are a widely used formalism for representing and reasoning about temporal constraints on activities. The dispatchability of an STN was originally defined as a guarantee that a specific real-time execution algorithm would necessarily satisfy all of the STN’s constraints while preserving maximum flexibility but requiring minimal computation. A Simple Temporal Network with Uncertainty (STNU) augments an STN to accommodate actions with uncertain durations. However, the dispatchability of an STNU was defined differently: in terms of the dispatchability of its so-called STN projections. It was then argued informally that this definition provided a similar real-time execution guarantee, but without specifying the execution algorithm. This paper formally defines a real-time execution algorithm for STNUs that similarly preserves maximum flexibility while requiring minimal computation. It then proves that an STNU is dispatchable if and only if every run of that real-time execution algorithm necessarily satisfies the STNU’s constraints no matter how the uncertain durations play out. By formally connecting STNU dispatchability to an explicit real-time execution algorithm, the paper fills in important elements of the foundations of the dispatchability of STNUs.
Download
Paper Nr: 139
Title:

Big Data Synthesis and Class Imbalance Rectification for Enhanced Forest Fire Classification Modeling
Authors:

Fatemeh Tavakoli, Kshirasagar Naik, Marzia Zaman, Richard Purcell, Srinivas Sampalli, Abdul Mutakabbir, Chung-Horng Lung and Thambirajah Ravichandran
Abstract: Forest fires have been escalating in frequency and intensity across Canada in recent times. This study employs machine learning techniques and builds a dataset framework utilizing Copernicus climate reanalysis data combined with historical fire data to develop a fire classification framework. Three algorithms, Random Forest, XGBoost, and LightGBM, were evaluated. Given the pronounced class imbalance of 154:1 between “non-fire” and “fire” events, we rigorously employed two re-sampling strategies: Spatiotemporal, focusing on spatial and seasonal considerations, and Technique-Driven, leveraging advanced algorithmic approaches. Ultimately, XGBoost combined with NearMiss Version 3 in a 0.09 sampling ratio between “non-fire” and “fire” events yielded the best results: 98.08% precision, 86.06% sensitivity, and 93.03% specificity.
Download
Paper Nr: 151
Title:

A Hierarchical Anytime k-NN Classifier for Large-Scale High-Speed Data Streams
Authors:

Aarti, Jagat S. Challa, Hrishikesh Harsh, Utkarsh D., Mansi Agarwal, Raghav Chaudhary, Navneet Goyal and Poonam Goyal
Abstract: The k-Nearest Neighbor Classifier (k-NN) is a widely used classification technique used in data streams. However, traditional k-NN-based stream classification algorithms can’t handle varying inter-arrival rates of objects in the streams. Anytime algorithms are a class of algorithms that effectively handle data streams that have variable stream speed and trade execution time with the quality of results. In this paper, we introduce a novel anytime k-NN classification method for data streams namely, ANY-k-NN. This method employs a proposed hierarchical structure, the Any-NN-forest, as its classification model. The Any-NN-forest maintains a hierarchy of micro-clusters with different levels of granularity in its trees. This enables ANY-k-NN to effectively handle variable stream speeds and incrementally adapt its classification model using incoming labeled data. Moreover, it can efficiently manage large data streams as the model construction is less expensive. It is also capable of handling concept drift and class evolution. Additionally, this paper also presents ANY-MP-k-NN, a first-of-its-kind framework for anytime k-NN classification of multi-port data streams over distributed memory architectures. ANY-MP-k-NN can efficiently manage very large and high-speed data streams and deliver highly accurate classification results. The experimental findings confirm the superior performance of the proposed methods compared to the state-of-the-art in terms of classification accuracy.
Download
Paper Nr: 162
Title:

Neural Architecture Search for Bearing Fault Classification
Authors:

Edicson B. Diaz, Enrique Naredo, Nicolas M. Díaz, Douglas M. Dias, Maria B. Diaz, Susan Harnett and Conor Ryan
Abstract: In this research, we address bearing fault classification by evaluating three neural network models: 1D Con-volutional Neural Network (1D-CNN), CNN-Visual Geometry Group (CNN-VGG), and Long Short-Term Memory (LSTM). Utilizing vibration data, our approach incorporates data augmentation to address the limited availability of fault class data. A significant aspect of our methodology is the application of neural architecture search (NAS), which automates the evolution of network architectures, including hyperparameter tuning, significantly enhancing model training. Our use of early stopping strategies effectively prevents overfitting, ensuring robust model generalization. The results highlight the potential of integrating advanced machine learning models with NAS in bearing fault classification and suggest possibilities for further improvements, particularly in model differentiation for specific fault classes.
Download
Paper Nr: 174
Title:

A Probabilistic Approach for Detecting Real Concept Drift
Authors:

Sirvan Parasteh and Samira Sadaoui
Abstract: Concept Drift (CD) is a significant challenge in real-world data stream applications, as its presence requires predictive models to adapt to data-distribution changes over time. Our paper introduces a new algorithm, Probabilistic Real-Drift Detection (PRDD), designed to track and respond to CD based on its probabilistic definitions. PRDD utilizes the classifier’s prediction errors and confidence levels to detect specifically the Real CD. In an exhaustive empirical study involving 16 synthetic datasets with Abrupt and Gradual drifts, PRDD is compared to well-known CD detection methods. PRDD is highly performing and shows a time complexity of O(1) per datapoint, ensuring its computational efficiency in high-velocity environments.
Download
Paper Nr: 175
Title:

Optimization of Fuzzy Rule Induction Based on Decision Tree and Truth Table: A Case Study of Multi-Class Fault Diagnosis
Authors:

Abdelouadoud Kerarmi, Assia Kamal-Idrissi and Amal F. Seghrouchni
Abstract: Fuzzy Logic (FL) offers valuable advantages in multi-classification tasks, offering the capability to deal with imprecise and uncertain data for nuanced decision-making. However, generating precise fuzzy sets requires substantial effort and expertise. Also, the higher the number of rules in the FL system, the longer the model’s computational time is due to the combinatorial complexity. Thus, good data description, knowledge extraction/representation, and rule induction are crucial for developing an FL model. This paper addresses these challenges by proposing an Integrated Truth Table in Decision Tree-based FL model (ITTDTFL) that generates optimized fuzzy sets and rules. C4.5 DT is employed to extract optimized membership functions and rules using Truth Table (TT) by eliminating the redundancy of the rules. The final version of the rules is extracted from the TT and used in the FL model. We compare ITTDTFL with state-of-the-art models, including FU-RIA, RIPPER, and Decision-Tree-based FL. Experiments were conducted on real datasets of machine failure, evaluating the performances based on several factors, including the number of generated rules, accuracy, and computational time. The results demonstrate that the ITTDTFL model achieved the best performance, with an accuracy of 98.92%, less computational time outperforming the other models.
Download
Paper Nr: 180
Title:

Towards Efficient Quantum Anomaly Detection: One-Class SVMs Using Variable Subsampling and Randomized Measurements
Authors:

Michael Kölle, Afrae Ahouzi, Pascal Debus, Robert Müller, Daniëlle Schuman and Claudia Linnhoff-Popien
Abstract: Quantum computing, with its potential to enhance various machine learning tasks, allows significant advancements in kernel calculation and model precision. Utilizing the one-class Support Vector Machine alongside a quantum kernel, known for its classically challenging representational capacity, notable improvements in average precision compared to classical counterparts were observed in previous studies. Conventional calculations of these kernels, however, present a quadratic time complexity concerning data size, posing challenges in practical applications. To mitigate this, we explore two distinct approaches: utilizing randomized measurements to evaluate the quantum kernel and implementing the variable subsampling ensemble method, both targeting linear time complexity. Experimental results demonstrate a substantial reduction in training and inference times by up to 95% and 25% respectively, employing these methods. Although unstable, the average precision of randomized measurements discernibly surpasses that of the classical Radial Basis Function kernel, suggesting a promising direction for further research in scalable, efficient quantum computing applications in machine learning.
Download
Paper Nr: 184
Title:

On Handling Concept Drift, Calibration and Explainability in Non-Stationary Environments and Resources Limited Contexts
Authors:

Sara Kebir and Karim Tabia
Abstract: In many real-world applications, we face two important challenges: The shift in data distribution and the concept drift on the one hand, and on the other hand, the constraints of limited computational resources, particularly in the field of IoT and edge AI. Although both challenges have been well studied separately, it is rare to tackle these two challenges together. In this paper, we put ourselves in a context of limited resources and we address the problem of the concept and distribution shift not only to ensure a good level of accuracy over time, but also we study the impact that this could have on two complementary aspects which are the confidence/calibration of the model as well as the explainability of the predictions in this context. We first propose a global framework for this problem based on incremental learning, model calibration and lightweight explainability. In particular, we propose a solution to provide feature attributions in a context of limited resources. Finally, we empirically study the impact of incremental learning on model calibration and the quality of explanations.
Download
Paper Nr: 192
Title:

Scalable Prediction of Atomic Candidate OWL Class Axioms Using a Vector-Space Dimension Reduced Approach
Authors:

Ali Ballout, Célia da Costa Pereira and Andrea B. Tettamanzi
Abstract: Scoring candidate axioms or assessing their acceptability against known evidence is essential for automated schema induction and can also be valuable for knowledge graph validation. However, traditional methods for accurately scoring candidate axioms are often computationally and storage expensive, making them impractical for use with large knowledge graphs. In this work, we propose a scalable method to predict the scores of atomic candidate OWL class axioms of different types. The method relies on a semantic similarity measure derived from the ontological distance between concepts in a subsumption hierarchy, as well as feature ranking and selection for vector-space dimension reduction. We train a machine learning model using our reduced vector-space, encode new candidates as a vector, and predict their scores. Extensive tests that cover a range of ontologies of various sizes and multiple parameters and settings are carried out to investigate the effectiveness and scalability of the method.
Download
Paper Nr: 194
Title:

Dynamically Choosing the Number of Heads in Multi-Head Attention
Authors:

Fernando F. Duarte, Nuno Lau, Artur Pereira and Luís P. Reis
Abstract: Deep Learning agents are known to be very sensitive to their parameterization values. Attention-based Deep Reinforcement Learning agents further complicate this issue due to the additional parameterization associated to the computation of their attention function. One example of this concerns the number of attention heads to use when dealing with multi-head attention-based agents. Usually, these hyperparameters are set manually, which may be neither optimal nor efficient. This work addresses the issue of choosing the appropriate number of attention heads dynamically, by endowing the agent with a policy πh trained with policy gradient. At each timestep of agent-environment interaction, πh is responsible for choosing the most suitable number of attention heads according to the contextual memory of the agent. This dynamic parameterization is compared to a static parameterization in terms of performance. The role of πh is further assessed by providing additional analysis concerning the distribution of the number of attention heads throughout the training procedure and the course of the game. The Atari 2600 videogame benchmark was used to perform and validate all the experiments.
Download
Paper Nr: 213
Title:

Embryo Development Stage Onset Detection by Time Lapse Monitoring Based on Deep Learning
Authors:

Wided S. Miled, Sana Chtourou, Nozha Chakroun and Khadija K. Berjeb
Abstract: In Vitro Fertilisation (IVF) is a procedure used to overcome a range of fertility issues, giving many couples the chance of having a baby. Accurate selection of embryos with the highest implantation potentials is a necessary step toward enhancing the effectiveness of IVF. The detection and determination of pronuclei number during the early stages of embryo development in IVF treatments help embryologists with decision-making regarding valuable embryo selection for implantation. Current manual visual assessment is prone to observer subjectivity and is a long and difficult process. In this study, we build a CNN-LSTM deep learning model to automatically detect pronuclear-stage in IVF embryos, based on Time-Lapse Images (TLI) of their early development stages. The experimental results proved possible the automation of pronuclei determination as the proposed deep learning based method achieved a high accuracy of 85% in the detection of pronuclear-stage embryo.
Download
Paper Nr: 226
Title:

Diversifying Knowledge Enhancement of Biomedical Language Models Using Adapter Modules and Knowledge Graphs
Authors:

Juraj Vladika, Alexander Fichtl and Florian Matthes
Abstract: Recent advances in natural language processing (NLP) owe their success to pre-training language models on large amounts of unstructured data. Still, there is an increasing effort to combine the unstructured nature of LMs with structured knowledge and reasoning. Particularly in the rapidly evolving field of biomedical NLP, knowledge-enhanced language models (KELMs) have emerged as promising tools to bridge the gap between large language models and domain-specific knowledge, considering the available biomedical knowledge graphs (KGs) curated by experts over the decades. In this paper, we develop an approach that uses lightweight adapter modules to inject structured biomedical knowledge into pre-trained language models (PLMs). We use two large KGs, the biomedical knowledge system UMLS and the novel biochemical ontology OntoChem, with two prominent biomedical PLMs, PubMedBERT and BioLinkBERT. The approach includes partitioning knowledge graphs into smaller subgraphs, fine-tuning adapter modules for each subgraph, and combining the knowledge in a fusion layer. We test the performance on three downstream tasks: document classification, question answering, and natural language inference. We show that our methodology leads to performance improvements in several instances while keeping requirements in computing power low. Finally, we provide a detailed interpretation of the results and report valuable insights for future work.
Download
Paper Nr: 229
Title:

An Optimised Ensemble Approach for Multivariate Multi-Step Forecasts Using the Example of Flood Levels
Authors:

Michel Spils and Sven Tomforde
Abstract: Deep Learning methods have become increasingly popular for time-series forecasting in recent years. One common way of improving time-series forecasts is to use ensembles. By combining forecasts of different models, for example calculating the mean forecast, it is possible to get an ensemble that performs better than each single member. This paper suggests a method of aggregating ensemble forecasts using another neural network.The focus is on multivariate multi-step ahead forecasting. Experiments are done on 5 water levels at small to medium-sized rivers and show improvements on naive ensembles and single neural networks.
Download
Paper Nr: 238
Title:

Parameter-Free Undersampling for Multi-Label Data
Authors:

Sarbani Palit and Payel Sadhukhan
Abstract: This work presents a novel undersampling scheme to tackle the imbalance problem in multi-label datasets. We use the principles of the natural nearest neighborhood and follow a paradigm of label-specific undersam-pling. Natural-nearest neighborhood is a parameter-free principle. Our scheme’s novelty lies in exploring the parameter-optimization-free natural nearest neighborhood principles. The class imbalance problem is particularly challenging in a multi-label context, as the imbalance ratio and the majority-minority distributions vary from label to label. Consequently, the majority-minority class overlaps also vary across the labels. Working on this aspect, we propose a framework where a single natural neighbor search is sufficient to identify all the label-specific overlaps. Natural neighbor information is also used to find the key lattices of the majority class (which we do not undersample). The performance of the proposed method, NaNUML, indicates its ability to mitigate the class-imbalance issue in multi-label datasets to a considerable extent. We could also establish a statistically superior performance over other competing methods several times. An empirical study involving twelve real-world multi-label datasets, seven competing methods, and four evaluating metrics - shows that the proposed method effectively handles the class-imbalance issue in multi-label datasets. In this work, we have presented a novel label-specific undersampling scheme, NaNUML, for multi-label datasets. NaNUML is based on the parameter-free natural neighbor search and the key factor, neighborhood size ’k’ is determined without invoking any parameter optimization.
Download
Paper Nr: 241
Title:

Efficient and Flexible Topic Modeling Using Pretrained Embeddings and Bag of Sentences
Authors:

Johannes Schneider
Abstract: Pre-trained language models have led to a new state-of-the-art in many NLP tasks. However, for topic modeling, statistical generative models such as LDA are still prevalent, which do not easily allow incorporating contextual word vectors. They might yield topics that do not align well with human judgment. In this work, we propose a novel topic modeling and inference algorithm. We suggest a bag of sentences (BoS) approach using sentences as the unit of analysis. We leverage pre-trained sentence embeddings by combining generative process models and clustering. We derive a fast inference algorithm based on expectation maximization, hard assignments, and an annealing process. The evaluation shows that our method yields state-of-the art results with relatively little computational demands. Our method is also more flexible compared to prior works leveraging word embeddings, since it provides the possibility to customize topic-document distributions using priors. Code and data is at https://github.com/JohnTailor/BertSenClu.
Download
Paper Nr: 242
Title:

A Survey of Deep Learning: From Activations to Transformers
Authors:

Johannes Schneider and Michalis Vlachos
Abstract: Deep learning has made tremendous progress in the last decade. A key success factor is the large amount of architectures, layers, objectives, and optimization techniques. They include a myriad of variants related to attention, normalization, skip connections, transformers and self-supervised learning schemes – to name a few. We provide a comprehensive overview of the most important, recent works in these areas to those who already have a basic understanding of deep learning. We hope that a holistic and unified treatment of influential, recent works helps researchers to form new connections between diverse areas of deep learning. We identify and discuss multiple patterns that summarize the key strategies for many of the successful innovations over the last decade as well as works that can be seen as rising stars. We also include a discussion on recent commercially built, closed-source models such as OpenAI’s GPT-4 and Google’s PaLM 2.
Download
Paper Nr: 244
Title:

DGDNN: Decoupled Graph Diffusion Neural Network for Stock Movement Prediction
Authors:

Zinuo You, Zijian Shi, Hongbo Bo, John Cartlidge, Li Zhang and Yan Ge
Abstract: Forecasting future stock trends remains challenging for academia and industry due to stochastic inter-stock dynamics and hierarchical intra-stock dynamics influencing stock prices. In recent years, graph neural networks have achieved remarkable performance in this problem by formulating multiple stocks as graph-structured data. However, most of these approaches rely on artificially defined factors to construct static stock graphs, which fail to capture the intrinsic interdependencies between stocks that rapidly evolve. In addition, these methods often ignore the hierarchical features of the stocks and lose distinctive information within. In this work, we propose a novel graph learning approach implemented without expert knowledge to address these issues. First, our approach automatically constructs dynamic stock graphs by entropy-driven edge generation from a signal processing perspective. Then, we further learn task-optimal dependencies between stocks via a generalized graph diffusion process on constructed stock graphs. Last, a decoupled representation learning scheme is adopted to capture distinctive hierarchical intra-stock features. Experimental results demonstrate substantial improvements over state-of-the-art baselines on real-world datasets. Moreover, the ablation study and sensitivity study further illustrate the effectiveness of the proposed method in modeling the time-evolving inter-stock and intra-stock dynamics.
Download
Paper Nr: 246
Title:

Spiral Drawing Test and Explainable Convolutional Neural Networks for Parkinson’s Disease Detection
Authors:

Francesco Mercaldo, Luca Brunese, Mario Cesarelli, Fabio Martinelli and Antonella Santone
Abstract: There is no definitive test for Parkinson’s disease, and the rate of misdiagnosis, particularly when made by individuals without specialized training, is significantly elevated. The spiral drawing test is a clinical assessment tool used to evaluate fine motor skills, hand-eye coordination, and tremor in individuals, particularly those with neurological disorders such as Parkinson’s disease. In this test, a person is typically asked to trace or draw a spiral pattern on a piece of paper or a digital tablet. The test measures the smoothness and steadiness of their hand movements. Any irregularities or tremors in the drawn spiral can provide valuable information to healthcare professionals in diagnosing or monitoring conditions like Parkinson’s disease, essential tremors, or other movement disorders. In this paper, we provide a method aimed at automatically analyse spiral drawing tests to understand whether a subject is affected by Parkinson’s disease. We employ two different Convolu-tional Neural Networks: DenseNet and ResNet50, by obtaining an accuracy equal to 0.96 in the evaluation of a dataset composed of 3,991 spiral drawing tests, thus showing the effectiveness of the proposed method. Moreover, with the aim to provide a kind of explainability behind the model prediction, the proposed method is able to visualise, directly on the spiral drawing test image, the areas of the test image that from the model point of view are related to Parkinson’s disease.
Download
Paper Nr: 255
Title:

ALISE: An Automated Literature Screening Engine for Research
Authors:

Hendrik Roth and Carsten Lanquillon
Abstract: The screening process needs the most time of a literature review. An automated approach saves a lot of time, making it easier for researchers to review literature. Most current approaches do not consider the full text for screening, which can cause the exclusion of relevant papers. The Automated LIterature Screening Engine (ALISE) performs full-text screening based on a research question about the retrieved papers of the literature search. With an average of 61.87% nWSS and a median of 74.38% nWSS, ALISE can save time for reviewers but cannot be used without human screening afterwards. Furthermore, ALISE is sensitive to the given research question(s).
Download
Paper Nr: 269
Title:

Predicting Major Donor Prospects Using Machine Learning
Authors:

Greg Lee, Aishwarya V. Sathyamurthi and Mark Hobbs
Abstract: An important concern for many fundraising institutions is major gift fundraising. Major gifts are large gifts (typically $10,000+) and donors who give these gifts are called major donors. Depending upon the institution type, major gifts can constitute 80% of donation dollars. Thus, being able to predict who will give a major gift is crucial for fundraising institutions. We sought the most useful major donor prospect model by experimenting with 11 shallow and deep learning algorithms. A useful model discovers major donor prospects (i.e., false positives) without generating a similar number of false negatives, helping to preserve accuracy. The study also examined the impact of using different types of data, such as donation data exclusively, on the model’s utility. Notably, an LSTM-GRU model achieved a 92.2% accuracy rate with 110 false positive prospects and 40 false negatives for a religious fundraising institution. This model could assist major donor officers in identifying potential major donors. Similarly, for an education fundraising institution, an extra trees classifier was able to generate a major donor model with 92.5% accuracy, 71 false positives and 40 false negatives. False positives are prospects for fundraising institutions, providing major gift officers potential major donors.
Download
Paper Nr: 275
Title:

Depth-Enhanced 3D Deep Learning for Strawberry Detection and Widest Region Identification in Polytunnels
Authors:

Gabriel Lins Tenorio, Weria Khaksar and Wouter Caarls
Abstract: This paper presents an investigation into the use of 3D Deep Learning models for enhanced strawberry detection in polytunnels. We focus on two main tasks: firstly, fruit detection, comparing the standard MaskRCNN and an adapted version that integrates depth information (MaskRCNN-D), both capable of classifying strawberries based on their maturity (ripe, unripe) and health status (affected by disease or fungus); secondly, for the identification of the widest region of strawberries, we compare a contour-based algorithm with an enhanced version of the VGG-16 model. Our findings demonstrate that integrating depth data into the MaskRCNN-D results in up to a 13.7% improvement in mean Average Precision (mAP) from 0.81 to 0.92 across various strawberry test sets, including simulated ones, emphasizing the model’s effectiveness in both real-world and simulated agricultural scenarios. Furthermore, our end-to-end pipeline approach, which combines the fruit detection (MaskRCNN-D) and widest region identification models (enhanced VGG-16), shows a remarkably low localization error, achieving down to 11.3 pixels of Root Mean Square Error (RMSE) in a 224 × 224 strawberry cropped image. This pipeline integration, combining the strengths of both models, provides the most effective result, enabling their application in autonomous fruit monitoring systems.
Download
Paper Nr: 288
Title:

Learning Occlusions in Robotic Systems: How to Prevent Robots from Hiding Themselves
Authors:

Jakob Nazarenus, Simon Reichhuber, Manuel Amersdorfer, Lukas Elsner, Reinhard Koch, Sven Tomforde and Hossam Abbas
Abstract: In many applications, robotic systems are monitored via camera systems. This helps with monitoring automated production processes, anomaly detection, and the refinement of the estimated robot’s pose via optical tracking systems. While providing high precision and flexibility, the main limitation of such systems is their line-of-sight constraint. In this paper, we propose a lightweight solution for automatically learning this occluded space to provide continuously observable robot trajectories. This is achieved by an initial autonomous calibration procedure and subsequent training of a simple neural network. During operation, this network provides a prediction of the visibility status with a balanced accuracy of 90% as well as a gradient that leads the robot to a more well-observed area. The prediction and gradient computations run with sub-ms latency and allow for modular integration into existing dynamic trajectory-planning algorithms to ensure high visibility of the desired target.
Download
Paper Nr: 309
Title:

On Learning Bipolar Gradual Argumentation Semantics with Neural Networks
Authors:

Caren Al Anaissy, Sandeep Suntwal, Mihai Surdeanu and Srdjan Vesic
Abstract: Computational argumentation has evolved as a key area in artificial intelligence, used to analyze aspects of thinking, making decisions, and conversing. As a result, it is currently employed in a variety of real-world contexts, from legal reasoning to intelligence analysis. An argumentation framework is modelled as a graph where the nodes represent arguments and the edges of the graph represent relations (i.e., supports, attacks) between nodes. In this work, we investigate the ability of neural network methods to learn a gradual bipolar argumentation semantics, which allows for both supports and attacks. We begin by calculating the acceptability degrees for graph nodes. These scores are generated using Quantitative Argumentation Debate (QuAD) argumentation semantics. We apply this approach to two benchmark datasets: Twelve Angry Men and Debate-pedia. Using this data, we train and evaluate the performance of three benchmark architectures: Multilayer Perceptron (MLP), Graph Convolution Network (GCN), and Graph Attention Network (GAT) to learn the acceptability degree scores produced by the QuAD semantics. Our results show that these neural network methods can learn bipolar gradual argumentation semantics. The models trained on GCN architecture perform better than the other two architectures underscoring the importance of modelling argumentation graphs explicitly. Our software is publicly available at: https://github.com/clulab/icaart24-argumentation.
Download
Paper Nr: 321
Title:

A Coachable Parser of Natural Language Advice
Authors:

Christodoulos Ioannou and Loizos Michael
Abstract: We present a system for parsing advice offered by a human to a machine. The advice is given in the form of conditional sentences in natural language, and the system generates a logic-based (machine-readable) rep-resentation of the advice, as appropriate for use by the machine in a downstream task. The system utilizes a “white-box” knowledge-based translation policy, which can be acquired iteratively in a developmental manner through a coaching process. We showcase this coaching process by demonstrating how linguistic annotations of sentences can be combined, through simple logic-based expressions, to carry out the translation task.
Download
Paper Nr: 325
Title:

Multiple Relations Classification Using Imbalanced Predictions Adaptation
Authors:

Sakher K. Alqaaidi, Elika Bozorgi and Krzysztof J. Kochut
Abstract: The relation classification task assigns the proper semantic relation to a pair of subject and object entities; the task plays a crucial role in various text mining applications, such as knowledge graph construction and entities interaction discovery in biomedical text. Current relation classification models employ additional procedures to identify multiple relations in a single sentence. Furthermore, they overlook the imbalanced predictions pattern. The pattern arises from the presence of a few valid relations that need positive labeling in a relatively large predefined relations set. We propose a multiple relations classification model that tackles these issues through a customized output architecture and by exploiting additional input features. Our findings suggest that handling the imbalanced predictions leads to significant improvements, even on a modest training design. The results demonstrate superiority performance on benchmark datasets commonly used in relation classification. To the best of our knowledge, this work is the first that recognizes the imbalanced predictions within the relation classification task.
Download
Paper Nr: 327
Title:

EAPC: Emotion and Audio Prior Control Framework for the Emotional and Temporal Talking Face Generation
Authors:

Xuan-Nam Cao, Quoc-Huy Trinh, Quoc-Anh Do-Nguyen, Van-Son Ho, Hoai-Thuong Dang and Minh-Triet Tran
Abstract: Generating realistic talking faces from audio input is a challenging task with broad applications in fields such as film production, gaming, and virtual reality. Previous approaches, employing a two-stage process of converting audio to landmarks and then landmarks to a face, have shown promise in creating vivid videos. However, they still face challenges in maintaining consistency due to misconnections between information from the previous audio frame and the current audio frame, leading to the generation of unnatural landmarks. To address this issue, we propose EAPC, a framework that incorporates features from previous audio frames with the current audio feature and the current facial landmark. Additionally, we introduce the Dual-LSTM module to enhance emotion control. By doing so, our framework improves the temporal aspects and emotional information of the audio input, allowing our model to capture speech dynamics and produce more coherent animations. Extensive experiments demonstrate that our method can generate consistent landmarks, resulting in more realistic and synchronized faces, leading to the achievement of our competitive results with state-of-the-art methods. The implementation of our method will be made publicly available upon publication.
Download
Paper Nr: 331
Title:

Machine Learning-Based Optimization of E-Commerce Advertising Campaigns
Authors:

Arti Jha, Pratyut Sharma, Ritik Upmanyu, Yashvardhan Sharma and Kamlesh Tiwari
Abstract: E-commerce platforms facilitate the generation of advertisement campaigns by retailers for the purpose of promoting their products. Marketers need to generate demand for their products by means of online advertising (ad). Game theoretic and continuous experimentation feedback-based advertising optimization is imperative to enable efficient and effective advertising at scale. To address this, we propose a solution that utilizes machine learning and statistical techniques to optimize e-commerce ad campaigns, intending to create an optimal and targeted ad campaign strategy. The dataset utilized here is Amazon’s e-commerce dataset obtained from a prominent e-commerce firm. The proposed work examines these key approaches: For predicting profitability and campaign impressions, we implemented a model using the first approach, blending statistical techniques with machine-learning algorithms. The results provide a comparison between the algorithms, offering insights into the observed outcomes. In the second approach, we leverage the k-means clustering algorithm and Bayesian Information Criterion (BIC) technique to establish a correlation between keyword performance, campaign profitability, and bidding strategies. In the concluding approach, we introduce an innovative model that uses Joint Probability Distribution and Gaussian functions to determine the profitability of ad campaigns. This model generates multivariate-density graphs, enabling a comprehensive exploration to better comprehend and predict profitability, specifically in terms of Return on Ad Spend (ROAS). For example, we can now answer questions like: How do the profitability (ROAS) and awareness (%impression share) of a campaign change with variations in the budget? How do the profitability (ROAS) and awareness (%impression share) of a keyword change with different bid values? These insights provide valuable information for optimizing campaign performance and making informed decisions regarding budget allocation, bid adjustments, and overall campaign structure. The results offer practical insights for optimizing an ad campaign’s performance through developing effective and targeted strategies.
Download
Paper Nr: 351
Title:

I-AM-Bird: A Deep Learning Approach to Detect Amazonian Bird Species in Residential Environments
Authors:

Lucas F. Zampar and Clay Palmeira da Silva
Abstract: The Amazon presents several challenges, such as recognizing and monitoring its birdlife. It is known that bird records are shared by many bird watchers in citizen science initiatives, including by residents who observe birds feeding at their home feeders. In this context, the work proposed an approach based on deep learning to automatically detect species of Amazonian birds that frequent residential feeders. To this end, a data set consisting of 940 images captured by 3 webcams installed in a residential feeder was collected. In total, 1,836 birds of 5 species were recorded and annotated. Then, we used the dataset to train different configurations of the Faster R-CNN detector. Considering the IoU threshold at 50%, the best model achieved an mAP of 98.33%, an mean precision of 95.96%, and an mean recall of 98.82%. The results also allow us to drive future works to develop a monitoring system for these species in a citizen science initiative.
Download
Paper Nr: 354
Title:

Evaluating Data Augmentation Techniques for Coffee Leaf Disease Classification
Authors:

Adrian Gheorghiu, Iulian-Marius Tăiatu, Dumitru-Clementin Cercel, Iuliana Marin and Florin Pop
Abstract: The detection and classification of diseases in Robusta coffee leaves are essential to ensure that plants are healthy and the crop yield is kept high. However, this job requires extensive botanical knowledge and much wasted time. Therefore, this task and others similar to it have been extensively researched subjects in image classification. Regarding leaf disease classification, most approaches have used the more popular PlantVillage dataset while completely disregarding other datasets, like the Robusta Coffee Leaf (RoCoLe) dataset. As the RoCoLe dataset is imbalanced and does not have many samples, fine-tuning of pre-trained models and multiple augmentation techniques need to be used. The current paper uses the RoCoLe dataset and approaches based on deep learning for classifying coffee leaf diseases from images, incorporating the pix2pix model for segmentation and cycle-generative adversarial network (CycleGAN) for augmentation. Our study demonstrates the effectiveness of Transformer-based models, online augmentations, and CycleGAN augmentation in improving leaf disease classification. While synthetic data has limitations, it complements real data, enhancing model performance. These findings contribute to developing robust techniques for plant disease detection and classification.
Download
Paper Nr: 355
Title:

Second-Order Learning with Grounding Alignment: A Multimodal Reasoning Approach to Handle Unlabelled Data
Authors:

Arnab Barua, Mobyen U. Ahmed, Shaibal Barua, Shahina Begum and Andrea Giorgi
Abstract: Multimodal machine learning is a critical aspect in the development and advancement of AI systems. However, it encounters significant challenges while working with multimodal data, where one of the major issues is dealing with unlabelled multimodal data, which can hinder effective analysis. To address the challenge, this paper proposes a multimodal reasoning approach adopting second-order learning, incorporating grounding alignment and semi-supervised learning methods. The proposed approach illustrates using unlabelled vehicular telemetry data. During the process, features were extracted from unlabelled telemetry data using an autoencoder and then clustered and aligned with true labels of neurophysiological data to create labelled and unlabelled datasets. In the semi-supervised approach, the Random Forest (RF) and eXtreme Gradient Boosting (XGBoost) algorithms are applied to the labelled dataset, achieving a test accuracy of over 97%. These algorithms are then used to predict labels for the unlabelled dataset, which is later added to the labelled dataset to retrain the model. With the additional prior labelled data, both algorithms achieved a 99% test accuracy. Confidence in predictions for unlabelled data was validated using counting samples based on the prediction score and Bayesian probability. RF and XGBoost scored 91.26% and 97.87% in counting samples and 98.67% and 99.77% in Bayesian probability, respectively.
Download
Paper Nr: 356
Title:

Analysis of the Effectiveness of Large Language Models in Assessing Argumentative Writing and Generating Feedback
Authors:

Daisy C. Albuquerque da Silva, Carlos Eduardo de Mello and Ana B. Garcia
Abstract: This study examines the use of Large Language Models (LLMs) like GPT-4 in the evaluation of argumentative writing, particularly opinion articles authored by military school students. It explores the potential of LLMs to provide instant, personalized feedback across different writing stages and assesses their effectiveness compared to human evaluators. The study utilizes a detailed rubric to guide the LLM evaluation, focusing on competencies from topic choice to bibliographical references. Initial findings suggest that GPT-4 can consistently evaluate technical and structural aspects of writing, offering reliable feedback, especially in the References category. However, its conservative classification approach may underestimate article quality, indicating a need for human oversight. The study also uncovers GPT-4’s challenges with nuanced and contextual elements of opinion writing, evident from variability in precision and low recall in recognizing complete works. These findings highlight the evolving role of LLMs as supplementary tools in education that require integration with human judgment to enhance argumentative writing and critical thinking in academic settings.
Download
Paper Nr: 358
Title:

LIDL4Oliv: A Lightweight Incremental Deep Learning Model for Classifying Olive Diseases in Images
Authors:

Emna Guermazi, Afef Mdhaffar, Mohamed Jmaiel and Bernd Freisleben
Abstract: We present LIDL4Oliv, a novel lightweight incremental deep learning model for classifying olive diseases in images. LIDL4Oliv is first trained on a novel annotated dataset of images with complex background. Then, it learns from a large-scale deep learning model, following a knowledge distillation approach. Finally, LIDL4Oliv is successfully deployed as a cross-platform application on resource-limited mobile devices, such as smartphones. The deployed deep learning can detect olive leaves in images and classify their states as healthy or unhealthy, i.e., affected by one of the two diseases “Aculus Olearius” and “Peacock Spot”. Our mobile application supports the collection of real data during operation, i.e., the training dataset is continuously augmented by newly collected images of olive leaves. Furthermore, our deep learning model is retrained in a continuous manner, whenever a new set of data is collected. LIDL4Oliv follows an incremental update process. It does not ignore the knowledge of the previously deployed model, but it (1) incorporates the current weights of the deployed model and (2) employs fine-tuning and knowledge distillation to create an enhanced incremental lightweight deep learning model. Our conducted experiments show the impact of using our complex background dataset to improve the classification results. They demonstrate the effect of using knowledge distillation in enhancing the performance of the deployed model on resource-limited devices.
Download
Paper Nr: 360
Title:

Towards the Use of AI-Based Tools for Systematic Literature Review
Authors:

Lotfi Souifi, Nesrine Khabou, Ismael B. Rodriguez and Ahmed H. Kacem
Abstract: The constant growth in the number of published research studies and their rapid rate of publication creates a significant challenge in identifying relevant studies for unbiased systematic reviews. To address this challenge, artificial intelligence (AI) methods have been used since 2016 to improve the efficiency of scientific review and synthesis. Nevertheless, the growth in the number of AI-powered tools dedicated to processing text-based data has been remarkable since the introduction of generative pre-trained transformers by OpenAI in late 2022. Moreover, alongside this development, ChatGPT, a language model that provides a user-friendly chatbot interface, was introduced. The incorporation of this interactive feature has greatly enhanced the capability of developers and end-users alike to effectively utilize and access ChatGPT. This study aims to investigate the effectiveness of six AI-based tools namely Chatpdf, Pdf2gpt, Hipdf, SciSpace, Easy-peasy AI, and DocAnalyzer AI, developed utilizing ChatGPT technology. These tools will be evaluated in a specific scenario where they are automated to carry out a particular step within a Systematic Literature Review. Furthermore, the limitations associated with each tool will be analyzed, and strategies will be proposed to overcome them. Additionally, this study aims to provide recommendations for researchers who intend to incorporate these tools into their research processes.
Download
Paper Nr: 365
Title:

A Unified Conceptual Framework Integrating UML and RL for Efficient Reconfiguration Design
Authors:

Amen Ben Hadj Ali and Samir Ben Ahmed
Abstract: The problem of early exploration of various design choices to anticipate potential runtime changes at design time for complex and highly-dynamic Reconfigurable Control Systems (RCS), is still a real challenge for designers. This paper proposes a novel conceptual framework that integrates the benefits of UML-based modeling with Reinforcement Learning (RL) to overcome this difficulty. Our proposal exploits UML diagrams enriched with OCL constraints to describe the reconfiguration controller structure and dynamics using predefined reconfiguration knowledge. On the other hand, the reconfiguration controller is designed as a RL agent (Reinforcement Learning Reconfiguration Agent or RLRA) able to improve its knowledge through online exploration while running a Q-Learning algorithm. The design process we propose starts with an abstract UML-based specification of RCS. Then, a RL-based framework in Python language will be generated from UML/OCL models by applying a generation algorithm. Finally, the resulting framework will be run to allow the RLRA learning optimized reconfiguration policies and eventually improve first design specifications with learning feedback. The learning phase supports both offline and online learning and is based on a Q-Learning algorithm.
Download
Paper Nr: 367
Title:

Hybrid Mechanistic Neural Network Modelling of the Degree of Cure of Polymer Composite
Authors:

Samuel Sells and Jie Zhang
Abstract: A hybrid mechanistic/neural network model was developed for the industrial polymer composite curing process of a fibre-reinforced polymer composite. A hybrid model with parallel scheme and a hybrid model with the combination of series and parallel schemes were developed. It is found that the hybrid model with the combination of series and parallel schemes gives better performance. It is shown that the developed hybrid model is more accurate than its mechanistic and neural network counterparts in predicting the degree of cure based upon the temperature and time data. The hybrid model is 7.7% and 17.1% more accurate than the neural network model and the mechanistic model respectively in terms of sum of absolute errors.
Download
Paper Nr: 372
Title:

Examining Decision-Making in Air Traffic Control: Enhancing Transparency and Decision Support Through Machine Learning, Explanation, and Visualization: A Case Study
Authors:

Christophe Hurter, Augustin Degas, Arnaud Guibert, Maelan Poyer, Nicolas Durand, Alexandre Veyrie, Ana Ferreira, Nicola Cavagnetto, Stefano Bonelli, Mobyen U. Ahmed, Waleed Jmoona, Shaibal Barua, Shahina Begum, Giulia Cartocci, Gianluca Di Flumeri, Gianluca Borghini, Fabio Babiloni and Pietro Aricó
Abstract: Artificial Intelligence (AI) has recently made significant advancements and is now pervasive across various application domains. This holds true for Air Transportation as well, where AI is increasingly involved in decision-making processes. While these algorithms are designed to assist users in their daily tasks, they still face challenges related to acceptance and trustworthiness. Users often harbor doubts about the decisions proposed by AI, and in some cases, they may even oppose them. This is primarily because AI-generated decisions are often opaque, non-intuitive, and incompatible with human reasoning. Moreover, when AI is deployed in safety-critical contexts like Air Traffic Management (ATM), the individual decisions generated by AI models must be highly reliable for human operators. Understanding the behavior of the model and providing explanations for its results are essential requirements in every life-critical domain. In this scope, this project aimed to enhance transparency and explainability in AI algorithms within the Air Traffic Management domain. This article presents the results of the project’s validation conducted for a Conflict Detection and Resolution task involving 21 air traffic controllers (10 experts and 11 students) in En-Route position (i.e. hight altitude flight management). Through a controlled study incorporating three levels of explanation, we offer initial insights into the impact of providing additional explanations alongside a conflict resolution algorithm to improve decision-making. At a high level, our findings indicate that providing explanations is not always necessary, and our project sheds light on potential research directions for education and training purposes.
Download
Paper Nr: 392
Title:

Using Chatbot Technologies to Support Argumentation
Authors:

Luis H. Herbets de Sousa, Guilherme Trajano, Analúcia S. Morales, Stefan Sarkadi and Alison R. Panisson
Abstract: Chatbots are extensively used in modern times and are exhibiting increasingly intelligent behaviors. However, being relatively new technologies, there are significant demands for further advancement. Numerous possibilities for research exist to refine these technologies, including integration with other technologies, especially in the field of artificial intelligence (AI), which has received much attention and development. This study aims to explore the ability of chatbot technologies to classify arguments according to the reasoning patterns used to create them. As argumentation is a significant aspect of human intelligence, categorizing arguments according to various argumentation schemes (reasoning patterns) is a crucial step towards developing sophisticated human-computer interaction interfaces. This will enable agents (chatbots) to engage in more sophisticated interactions, such as argumentation processes.
Download
Paper Nr: 393
Title:

Spread and (Mis)use of Evaluative Expressions in Human Written and LLM-Based Generated Text
Authors:

Maurice Langner and Ralf Klabunde
Abstract: We investigate the capacity of Large Language Models (LLMs) to generate evaluative expressions in a data-driven manner. The linguistic object of investigation is the production of justified and adequate evaluative language, such that the evaluative stance of the text is motivated by the underlying data. We use the SportSett corpus for generating summaries of basketball games. The input data is converted into RDF triples that are fed into GPT-4 and GPT-3.5, prompting the models to produce game summaries using evaluative adverbs and judgemental language. We annotated the generated texts and the original summaries for their propositional content contained in the line score and box score of each game, as well as for evaluative adverbs and their polarity. The results show that the models struggle to correctly interpret the numerical data and coherently assess the quality of team-wise and player-wise performances both within games and across games, often producing contradictory evaluations and displaying the lack of global evaluative scales.
Download
Short Papers
Paper Nr: 13
Title:

Multiverse: A Deep Learning 4X4 Sudoku Solver
Authors:

Chaim Schendowich, Eyal Ben Isaac and Rina Azoulay
Abstract: This paper presents a novel deep learning-based approach to solving 4x4 Sudoku puzzles, by viewing Sudoku as a complex multi-level sequence completion problem. It introduces a neural network model, termed as ”Multiverse”, which comprises multiple parallel computational units, or ”verses”. Each unit is designed for sequence completion based on Long Short-Term Memory (LSTM) modules. The paper’s novel perspective views Sudoku as a sequence completion task rather than a pure constraint satisfaction problem. The study generated its own dataset for 4x4 Sudoku puzzles and proposed variants of the Multiverse model for comparison and validation purposes. Comparative analysis shows that the proposed model is competitive with, and potentially superior to, state-of-the-art models. Notably, the proposed model was able to solve the puzzles in a single prediction, which offers promising avenues for further research on larger, more complex Sudoku puzzles.
Download
Paper Nr: 15
Title:

Which Objective Function is Solved Faster in Multi-Agent Pathfinding? It Depends
Authors:

Jiří Švancara, Dor Atzmon, Klaus Strauch, Roland Kaminski and Torsten Schaub
Abstract: Multi-agent pathfinding (MAPF) is the problem of finding safe paths for multiple mobile agents within a shared environment. This problem finds practical applications in real-world scenarios like navigation, warehousing, video games, and autonomous intersections. Finding the optimal solution to MAPF is known to be computationally hard. In the literature, two commonly used cost functions are makespan and the sum of costs. To tackle this complex problem, various algorithms have been developed, falling into two main categories: search-based approaches (e.g., Conflict Based Search) and reduction-based approaches, including reduction to SAT or ASP. In this study, we empirically compare these two approaches in the context of both makespan and the sum of costs, aiming to identify situations where one cost function presents more challenges than the other. We compare our results with older studies and improve upon their findings. Despite these solving approaches initially being designed for different cost functions, we observe similarities in their behavior. Furthermore, we identify a tipping point related to the size of the environment. On smaller maps, the sum of costs is more challenging, while makespan poses greater difficulties on larger maps for both solving paradigms, defying intuitive expectations. Our study also offers insights into the reasons behind this behavior.
Download
Paper Nr: 16
Title:

Variance Reduction of Resampling for Sequential Monte Carlo
Authors:

Xiongming Dai and Gerald Baumgartner
Abstract: A resampling scheme provides a way to switch low-weight particles for sequential Monte Carlo with higherweight particles representing the objective distribution. The less the variance of the weight distribution is, the more concentrated the effective particles are, and the quicker and more accurate it is to approximate the hidden Markov model, especially for the nonlinear case. Normally the distribution of these particles is skewed, we propose repetitive ergodicity in the deterministic domain with the median for resampling and have achieved the lowest variances compared to the other resampling methods. As the size of the deterministic domain M ≪ N (the size of population), given a feasible size of particles under mild assumptions, our algorithm is faster than the state of the art, which is verified by theoretical deduction and experiments of a hidden Markov model in both the linear and non-linear cases.
Download
Paper Nr: 19
Title:

A Semi-Automatic Light-Weight Approach Towards Data Generation for a Domain-Specific FAQ Chatbot Using Human-in-the-Loop
Authors:

Anum Afzal, Tao Xiang and Florian Matthes
Abstract: Employees at large companies tend to have longer waiting times if they need company-specific information and similarly someone on the other end needs to manually address those queries. Most companies are trying to incorporate LLM-powered conversational agents to make this processing faster but often struggle to find appropriate training data, especially domain-specific data. This paper introduces a semi-automatic approach for generating domain-specific training data while leveraging a domain-expert as a human-in-the-loop for quality control. We test this approach on a HR use-case of a large organization through a retrieval-based question-answering pipeline. Additionally, we also test the effect of long context on the performance of the FAQ chat for which we employ LongT5, an Efficient Transformer. Our experiments using LongT5 show that the inclusion of the generated training data improves the performance of the FAQ chatbot during inference.
Download
Paper Nr: 23
Title:

Conjugate Gradient for Latent Space Manipulation
Authors:

Walid Messaoud, Rim Trabelsi, Adnane Cabani and Fatma Abdelkefi
Abstract: Generative Adversarial Networks (GANs) have revolutionized image generation, allowing the production of high-quality images from latent codes in the latent space. However, manipulating the latent space to achieve specific image attributes remains challenging. Existing methods often lack disentanglement, leading to unintended changes in other attributes. Moreover, most of the existing techniques are limited to one-dimensional conditioning, making them less effective for complex multidimensional modifications. In this paper, we propose a novel approach that combines an auxiliary map composed of convolutional layers and Conjugate Gradient (CG) to enhance latent space manipulation. The proposed auxiliary map provides a versatile and expressive way to incorporate external information for image generation, while CG facilitates precise and controlled manipulations. Our experimental results demonstrate better performance compared to state-of-the-art methods.
Download
Paper Nr: 30
Title:

Autonomous Drone Takeoff and Navigation Using Reinforcement Learning
Authors:

Sana Ikli and Ilhem Quenel
Abstract: Unmanned Aerial Vehicles, also known as drones, are deployed in various applications such as security and surveillance. They also have the key benefit of being able to operate without a human pilot, which make them suitable to access difficult areas. During autonomous flights, drones can crash or collide with an obstacle. To prevent such situation, they need an obstacle-avoidance solution. In this work, we are interested in the navigation with obstacle avoidance of a single drone. The goal is to autonomously navigate from an origin to a destination point, including takeoff, without crashing. Reinforcement learning is a valid solution to this problem. Indeed, these approaches, coupled with deep learning, are used to tackle complex problems in robotics. However, the works in the literature using reinforcement learning for drone navigation usually simplify the problem into 2-D navigation. We propose to extend these approaches to complete 3-D navigation by using a state-of-the-art algorithm: proximal policy optimization. To create realistic drone environments, we will use a 3-D simulator called Pybullet. Results show that the drone successfully takes off and navigates to the indicated point. We provide in this paper a link to our video demonstration of the drone performing navigation tasks.
Download
Paper Nr: 33
Title:

A Study on Drug Similarity Measures for Predicting Drug-Drug Interactions and Severity Using Machine Learning Techniques
Authors:

Deepa Kumari
Abstract: Drug-Drug interaction (DDI) can lead to adverse reactions by decreasing the absorption rate in a patient body. The existing literature has limited focus on the impact of various similarity measures on DDI effects. This paper analyzes seven drug features (chemical substructures, targets, transporters, enzymes, side-effects, offsides, and carriers) obtained from Drugbank, Sider, TWOSIDES, and OFFSIDE databases to analyze DDI. This research examines five Machine Learning models (Logistic Regression, Random Forest, Decision Tree, KNN, ANN) on 16 different similarity measures to observe the performance of predicting samples through accuracy and AUC-curve analysis. The Jaccard similarity is chosen for further DDI prediction as it gives the best similarity score. The feature selection process (using Chi-Square) further reduces the time and space complexity. It compares combinations of every selected feature (chemical substructures, side-effects, offsides, enzymes) on Logistic Regression, Random Forest, and XGB classifiers. The results show that the Random Forest Classifier predicts DDI with the best accuracy of 72%. It also uniquely categorizes the severity level of side effects (minor, moderate, and major) due to DDI events through multi-class classification. Thus, it gives a better clinical significance to fast-track the clinical trials.
Download
Paper Nr: 37
Title:

Activity Recognition in Smartphones Using Non-Intrusive Sensors
Authors:

Pedro Fernandes, Cesar Analide and Bruno Fernandes
Abstract: Activity recognition using smartphones has gained increased attention in recent years due to the widespread adoption of these devices and, consequently, their various sensors. These sensors are capable of providing very relevant data for this purpose. Non-intrusive sensors, in particular, offer the advantage of collecting data without requiring the user to perform any specific action or use any additional devices. The objective of this study was, therefore, the development of an application designed for activity recognition using exclusively non-intrusive sensors available in any smartphone. The data collected by these sensors underwent several processing stages, and after numerous iterations, a set of highly favorable features for training the machine learning models was obtained. The most prominent result was achieved by the model using the XGBoost algorithm, which achieved an impressive accuracy rate of 0.979. This quite robust result confirms the high effectiveness of using this type of sensors for activity recognition.
Download
Paper Nr: 38
Title:

Automatic Detection and Classification of Atmospherical Fronts
Authors:

Andreea A. Ploscar, Anca I. Muscalagiu, Eduard T. Pauliuc and Adriana M. Coroiu
Abstract: This paper presents an application that uses Convolutional Neural Networks (CNN) for the automatic detection and classification of atmospherical fronts in synoptic maps, which are a graphical representation of weather conditions over a specific geographic area at a given point in time. These fronts are significant indicators of meteorological characteristics and are essential for weather forecasting. The proposed method takes in a region extracted from a synoptic map to detect and classify fronts as cold, warm, or mixed, setting our study apart from existing literature. Furthermore, unlike previous research that typically utilizes atmospheric data grids, our study employs synoptic maps as input data. Additionally, our model produces a single output, accurately representing the front type with a 78% accuracy rate. The CNN model was trained on data collected from various meteorological stations worldwide between 2013 and 2022. The proposed tool can provide valuable information to weather forecasters and improve their accuracy.
Download
Paper Nr: 44
Title:

A Formal Verification Approach to Handle Attack Graphs
Authors:

Davide Catta, Jean Leneutre, Antonina Mijatovic, Johanna Ulin and Vadim Malvone
Abstract: We propose a formalization of attack graphs through a multi-agent approach. Specifically, we focus on dynamic scenarios that capture the interaction between an attacker and defenders during a cyberattack. We introduce a formal definition of an attack graph using interpreted systems, demonstrating how this formalization enables us to express interesting security properties. Finally, we present a tool AG2IS, which we have developed as an implementation of our formal definitions, to perform the formal verification of attack graphs.
Download
Paper Nr: 46
Title:

Requisite Social Influence in Self-Regulated Systems
Authors:

Asimina Mertzani and Jeremy Pitt
Abstract: This paper specifies, implements and experiments with a new psychologically-inspired 4voices algorithm to be used by the units of a self-regulated system, whereby each unit learns to identify which of several “voices” to pay attention to, depending on a collective desired outcome (e.g., establishing the ground truth, a community truth, or their own “truth”). In addition, a regulator uses a standard Q-learning algorithm to pay attention to the regulated units and respond accordingly. The algorithm is applied to a problem of continuous policy-based monitoring and control, and simulation experiments determine which initial conditions produce systemic stability and what kind of “truth” is expressed by the regulated units. We conclude that this synthesis of Q-learning in the regulator and 4voices in the regulated system establishes requisite social influence . This maintains quasi-stability (i.e. periodic stability) and points the way towards ethical regulators.
Download
Paper Nr: 50
Title:

A Novel Metric for Measuring Data Quality in Classification Applications
Authors:

Jouseau Roxane, Salva Sébastien and Samir Chafik
Abstract: Data quality is a key element for building and optimizing good learning models. Despite many attempts to characterize data quality, there is still a need for rigorous formalization and an efficient measure of the quality from available observations. Indeed, without a clear understanding of the training and testing processes, it is hard to evaluate the intrinsic performance of a model. Besides, tools allowing to measure data quality specific to machine learning are still lacking. In this paper, we introduce and explain a novel metric to measure data quality. This metric is based on the correlated evolution between the classification performance and the deterioration of data. The proposed method has the major advantage of being model-independent. Furthermore, we provide an interpretation of each criterion and examples of assessment levels. We confirm the utility of the proposed metric with intensive numerical experiments and detail some illustrative cases with controlled and interpretable qualities.
Download
Paper Nr: 51
Title:

Solving Job Shop Problems with Neural Monte Carlo Tree Search
Authors:

Marco Kemmerling, Anas Abdelrazeq and Robert H. Schmitt
Abstract: Job shop scheduling is a common NP-hard problem that finds many applications in manufacturing and beyond. A variety of methods to solve job shop problems exist to address different requirements arising from individual use cases. Recently, model-free reinforcement learning is increasingly receiving attention as a method to train agents capable of scheduling. In contrast, model-based reinforcement learning is less well studied in job scheduling. However, it may be able to improve upon its model-free counterpart by dynamically spending additional planning budget to refine solutions according to the available scheduling time at any given moment. Neural Monte Carlo tree search, a family of model-based algorithms including AlphaZero is especially suitable for discrete problems such as the job shop problem. Our aim is to find suitable designs of neural Monte Carlo tree search agents for the job shop problem by systematically varying certain parameters and design components. We find that different choices for the evaluation phase of the tree search have the biggest impact on performance and conclude that agents with a combination of node value initialization using learned value functions and roll-out based evaluation lead to the most favorable performance.
Download
Paper Nr: 52
Title:

DiT-Head: High Resolution Talking Head Synthesis Using Diffusion Transformers
Authors:

Aaron Mir, Eduardo Alonso and Esther Mondragón
Abstract: We propose a novel talking head synthesis pipeline called ”DiT-Head,” which is based on diffusion transformers and uses audio as a condition to drive the denoising process of a diffusion model. Our method is scalable and can generalise to multiple identities while producing high-quality results. We train and evaluate our proposed approach and compare against existing methods of talking head synthesis. We show that our model can compete with these methods in terms of visual quality and lip-sync accuracy. Our results highlight the potential of our proposed approach to be used for a wide range of applications including virtual assistants, entertainment, and education. For a video demonstration of results and our user study, please refer to our supplementary material.
Download
Paper Nr: 55
Title:

Contextual Online Imitation Learning (COIL): Using Guide Policies in Reinforcement Learning
Authors:

Alexander Hill, Marc Groefsema, Matthia Sabatelli, Raffaella Carloni and Marco Grzegorczyk
Abstract: This paper proposes a novel method of utilising guide policies in Reinforcement Learning problems; Contextual Online Imitation Learning (COIL). This paper demonstrates that COIL can offer improved performance over both offline Imitation Learning methods such as Behavioral Cloning, and also Reinforcement Learning algorithms such as Proximal Policy Optimisation which do not take advantage of existing guide policies. An important characteristic of COIL is that it can effectively utilise guide policies that exhibit expert behavior in only a strict subset of the state space, making it more flexible than classical methods of Imitation Learning. This paper demonstrates that through using COIL, guide policies that achieve good performance in sub-tasks can also be used to help Reinforcement Learning agents looking to solve more complex tasks. This is a significant improvement in flexibility over traditional Imitation Learning methods. After introducing the theory and motivation behind COIL, this paper tests the effectiveness of COIL on the task of mobile-robot navigation in both a simulation and real-life lab experiments. In both settings, COIL gives stronger results than offline Imitation Learning, Reinforcement Learning, and also the guide policy itself.
Download
Paper Nr: 56
Title:

Exploring Patterns and Assessing the Security of Pseudorandom Number Generators with Machine Learning
Authors:

Sara Boancă
Abstract: In recent years, Machine Learning methods have been employed for testing the security of pseudorandom number generators. It is considered that successful learning from pseudorandom data implies the existence of some detectable pattern within it, thus reducing the generator security. As the number and complexity of such approaches has reported important growth, the aim of the present paper is to synthesize current results, discuss perspectives and challenges and provide relevant guidelines for future study. To the best of our knowledge, this is the first comprehensive analysis on the current state of the research into the problem of pseudorandomness exploration by means of Machine Learning.
Download
Paper Nr: 62
Title:

Solving Many-Objective Optimization Problems Using Selection Hyper-Heuristics
Authors:

Adeem Ali Anwar, Guanfeng Liu and Xuyun Zhang
Abstract: To effectively solve discrete optimization problems, meta-heuristics and heuristics have been used but their performance suffers drastically in the cross-domain applications. Hence, hyper-heuristics (HHs) have been used to cater to cross-domain problems. In literature, different HHs and meta-heuristics have been applied to solve the Many-objective Job-Shop Scheduling problem (MaOJSSP) and Many-objective Knapsack problem (MaOKSP) but the results are not convincing. Furthermore, no researchers have tried to solve these problems as cross-domain together using HHs. Additionally, the considered HH known as the cricket-based selection hyper-heuristic (CB-SHH) has not applied to any variation of the Job-shop scheduling problem (JSP) and the knapsack problem (KSP). This paper compares the performance of recently proposed HHs named CB-SHH, H-ACO, MARP-NSGAIII, and meta-heuristics named MPMOGA, MOEA/D on MaOKSP, MaOJSSP and benchmark problems. The performance of state-of-the-art HHs and meta-heuristics have been compared using hypervolume (HV) and µ norm. The main contribution of the paper is to effectively solve the MaOJSSP and MaOKSP using HHs and to prove the effectiveness of the best HHs on benchmark problems. It is proven through experiments that the CB-SHH is the best-performing algorithm on 44 out of 48 instances across all datasets and is the best cross-domain algorithm across the datasets.
Download
Paper Nr: 63
Title:

GenGradAttack: Efficient and Robust Targeted Adversarial Attacks Using Genetic Algorithms and Gradient-Based Fine-Tuning
Authors:

Naman Agarwal and James Pope
Abstract: Adversarial attacks pose a critical threat to the reliability of machine learning models, potentially undermining trust in practical applications. As machine learning models find deployment in vital domains like au-tonomous vehicles, healthcare, and finance, they become susceptible to adversarial examples—crafted inputs that induce erroneous high-confidence predictions. These attacks fall into two main categories: white-box, with full knowledge of model architecture, and black-box, with limited or no access to internal details. This paper introduces a novel approach for targeted adversarial attacks in black-box scenarios. By combining genetic algorithms and gradient-based fine-tuning, our method efficiently explores input space for perturbations without requiring access to internal model details. Subsequently, gradient-based fine-tuning optimizes these perturbations, aligning them with the target model’s decision boundary. This dual strategy aims to evolve perturbations that effectively mislead target models while minimizing queries, ensuring stealthy attacks. Results demonstrate the efficacy of GenGradAttack, achieving a remarkable 95.06% Adversarial Success Rate (ASR) on MNIST with a median query count of 556. In contrast, conventional GenAttack achieved 100% ASR but required significantly more queries. When applied to InceptionV3 and Ens4AdvInceptionV3 on ImageNet, GenGradAttack outperformed GenAttack with 100% and 96% ASR, respectively, and fewer median queries. These results highlight the efficiency and effectiveness of our approach in generating adversarial examples with reduced query counts, advancing our understanding of adversarial vulnerabilities in practical contexts.
Download
Paper Nr: 64
Title:

Cryptocurrency Analysis: Price Prediction of Cryptocurrency Using User Sentiments and Quantitative Data
Authors:

Dayan Perera, Jessica Lim, Shuta Gunraku and Wern H. Lim
Abstract: This research introduces an innovative approach to forecasting cryptocurrency prices by combining user-generated content (UGC) and sentiment analysis with quantitative data. The primary goal is to overcome limitations in existing methods for market forecasting, where accurate forecasting is crucial for informed decision-making and risk mitigation. The paper suggests a robust prediction methodology by integrating sentiment analysis and quantitative data. The study reviews prior research on sentiment analysis and quantitative analysis of cryptocurrency and stock price prediction. It explores the integration of machine learning and deep learning techniques, an area not extensively explored before. The methodology employs Long Short-Term Memory (LSTM), Recurrent Neural Network (RNN), Bidirectional LSTM and Gated Recurrent Unit (GRU) models to capture temporal dependencies. Prediction accuracy is assessed using metrics including Mean Squared Error (MSE), Root Mean Squared Error (RMSE), and a confusion matrix. Results show that GRU models excel in prediction, while RNN models outperform in predicting price movements; with an emphasis on the significance of a suitable data preprocessing pipeline towards improving model performance. In summary, this study demonstrates the effectiveness of integrating sentiment analysis and quantitative data for cryptocurrency price forecasting using UGC data.
Download
Paper Nr: 71
Title:

Outlier Detection in MET Data Using Subspace Outlier Detection Method
Authors:

Dupuy R. Charles, Pascal Pultrini and Andrea Tettamanzi
Abstract: In plant breeding, Multi-Environment Field Trials (MET) are commonly used to evaluate genotypes for multiple traits and to estimate their genetic breeding value using Genomic Prediction (GP). The occurrence of outliers in MET is common and is known to have a negative impact on the accuracy of the GP. Therefore, identification of outliers in MET prior to GP analysis can lead to better results. However, Outlier Detection (OD) in MET is often overlooked. Indeed, MET give rise to different level of residuals which favor the presence of swamping and masking effects where ideal sample points may be portrayed as outliers instead of the true ones. Consequently, without a sensitive and robust outlier detection algorithm, OD can be a waste of time and potentially degrade the accuracy prediction of the GP, especially when the data set is not huge. In this study, we compared various robust outlier methods from different approaches to determine which one is most suitable for identifying MET anomalies. Each method has been tested on eleven real-world MET data sets. Results are validated by injecting a proportion of artificial outliers in each set. The Subspace Outlier Detection Method stands out as the most promising among the tested methods.
Download
Paper Nr: 72
Title:

Hybrid Approach to Explain BERT Model: Sentiment Analysis Case
Authors:

Aroua Hedhili and Islem Bouallagui
Abstract: The increasing use of Artificial Intelligence (AI), particularly Deep Neural Networks (DNNs), has raised concerns about the opacity of these ’black box’ models in decision-making processes. Explainable AI (XAI) has emerged to address this issue by making AI systems more understandable and trustworthy through various techniques. In this research paper, we deal with a new approach to explain model combining counterfactual explanations and domain knowledge visualization. Our contribution explores how domain knowledge, guided by expert decision-makers, can improve the effectiveness of counterfactual explanations. Additionally, the presented research underscores the significance of collecting user feedback to create a human-centered approach. Our experiments were conducted on a BERT model for sentiment analysis on IMDB movie reviews dataset.
Download
Paper Nr: 76
Title:

Joining LDA and Word Embeddings for Covid-19 Topic Modeling on English and Arabic Data
Authors:

Amina Amara, Mohamed Ali Hadj Taieb and Mohamed Ben Aouicha
Abstract: The value of user-generated content on social media platforms has been well established and acknowledged since their rich and subjective information allows for favorable computational analysis. Nevertheless, social data are often text-heavy and unstructured, thereby complicating the process of data analysis. Topic models act as a bridge between social science and unstructured social data analysis to provide new perspectives for interpreting social phenomena. Latent Dirichlet Allocation (LDA) is one of the most used topic modeling techniques. However, the LDA-based topic models alone do not always provide promising results and do not consider the recent advancement in the natural language processing field by leveraging word embeddings when learning latent topics to capture more word-level semantic and syntactic regularities. In this work, we extend the LDA model by mixing the Skip-gram model with Dirichlet-optimized sparse topic mixtures to learn dense word embeddings jointly with the Dirichlet distributed latent document-level mixtures of topic vectors. The embeddings produced through the proposed model were submitted to experimental evaluation using a Covid-19 based multilingual dataset extracted from the Facebook social network. Experimental results show that the proposed model outperforms all compared baselines in terms of both topic quality and predictive performance.
Download
Paper Nr: 82
Title:

An Approach for Improving Oversampling by Filtering out Unrealistic Synthetic Data
Authors:

Nada Boudegzdame, Karima Sedki, Rosy Tspora and Jean-Baptiste Lamy
Abstract: Oversampling algorithms are commonly used in machine learning to address class imbalance by generating new synthetic samples of the minority class. While oversampling can improve classification models’ performance on minority classes, our research reveals that models often learn to detect noise generated by oversampling algorithms rather than the underlying patterns. To overcome this issue, this article proposes a method that involves identifying and filtering unrealistic synthetic data, using advanced technique such a neural network for detecting unrealistic synthetic data samples. This aims to enhance the quality of the oversampled datasets and improve machine learning models’ ability to uncover genuine patterns. The effectiveness of the proposed approach is thoroughly examined and evaluated, demonstrating enhanced model performance.
Download
Paper Nr: 94
Title:

Why an Automated, Scalable and Resilient Service for Semantic Interoperability is Needed
Authors:

Maximilian Stäbler, Tobias Müller, Frank Köster and Christoph Schlueter-Langdon
Abstract: The increasing linkage of different data sources and data ecosystems underlines the need for high-quality and well-structured data. Unambiguous descriptions of data (meta-data) promote a common understanding of the data among different users. New ontologies and data schemas are constantly being developed for this purpose. While there are new ways to align, merge or match these ontologies and data schemas, the context of the data, which is important for a clear understanding, is often not taken into account. This work addresses this problem by analyzing a graph consisting of 1,615 data attributes from 13 domains and 828 different ontologies. The results show how overlapping and partially synonymous ontologies, both from the same domain and from different domains, are. The results show the complexity for users in creating unique descriptions of data and why new approaches and methods are needed to achieve semantic interoperability.
Download
Paper Nr: 95
Title:

Comparative Analysis of Internal and External Facial Features for Enhanced Deep Fake Detection
Authors:

Fatimah Alanazi
Abstract: In the burgeoning era of deepfake technologies, the authenticity of digital media is being perpetually challenged, raising pivotal concerns regarding its veracity and the potential malicious uses of manipulated content. This study embarks on a meticulous exploration of the effectiveness of both internal and external facial features in discerning deepfake content. By conducting a thorough comparative analysis, our research illuminates the criticality of facial features, particularly those situated beyond the face’s center, in distinguishing between genuine and manipulated faces. The results elucidate that such features serve as potent indicators, thereby offering valuable insights for enhancing deepfake detection methodologies. Consequently, this research, therefore, not only underscores the paramount importance of these often-overlooked facial aspects but also contributes substantively to the domain of digital forensics, providing a nuanced understanding and innovative approaches towards advancing deepfake detection strategies. By bridging the gap between technological advancements and ethical digital media practices, this study stands as a beacon, advocating for the imperative need to safeguard the integrity of digital communications in our progressively digitized world.
Download
Paper Nr: 96
Title:

Seeing Through the Smoke: An Agent Architecture for Representing Health Protection Motivation Under Social Pressure
Authors:

Veronika Kurchyna, Stephanie C. Rodermund, Ye Eun Bae, Patrick Mertes, Philipp Flügger, Jan Ole Berndt and Ingo J. Timm
Abstract: Representing and emulating human decision-making processes in artificial intelligence systems is a challenging task. This is because both internal (such as attitude, perceived health or motivation) and external factors (such as the opinions of others) and their mutual interactions affect decision-making. Modelling agents capable of human-like behavior, including undesirable actions, is an interesting use case for designing different AI-systems when it comes to human-AI-interactions and similar scenarios. However, agent-based decision-models in this domain tend to reflect the complex interplay of these factors only to a limited extent. To overcome this, we enrich these approaches with an agent architecture inspired by theories from psychology and sociology. Using human health behavior, specifically smoking, as a case study, we propose an agent-based approach to combine social pressure within Protection Motivation Theory (PMT) to allow for a theory-based representation of potentially harmful behavior including both internal and external factors. Based on smoking in social settings, we present experiments to demonstrate the model’s capability to simulate human health behavior and the mutual influences between the selected concepts. In this use case, the resulting model has shown that social pressure is a driving influence in the observable system dynamics.
Download
Paper Nr: 97
Title:

Spatial-Temporal Graph Neural Network for the Detection of Container Escape Events
Authors:

Yuchen Guo and James Pope
Abstract: Internet of Things (IoT) devices bring an attack surface closer to personal life and industrial production. With containers as the primary method of IoT application deployment, detecting container escapes by analyzing audit logs can identify compromised edge devices. Since audit log data contains temporal property of events and relational information between system entities, existing analysis methods cannot comprehensively analyze these two properties. In this paper, a new Temporal Graph Neural Network (GNN) -based model was designed to detect anomalies of IoT applications in a container environment. The model employed Gated Recurrent Unit (GRU) and Graph Isomorphism Network (GIN) operators to capture temporal and spatial features. Using unsupervised learning to model the application’s normal behavior, the model can detect unknown anomalies that have not appeared in training. The model is trained on a dynamic graph generated from audit logs, which records security events in a system. Due to the lack of real-world datasets, we conducted experiments on a simulated dataset. Audit log records are divided into multiple graphs according to their temporal attribute to form a dynamic graph. Some nodes and edges are aggregated or removed to reduce the complexity of the graph. In the Experiments, The model has an F1 score of 0.976 on the validation set, which outperforms the best-performing baseline model, with an F1 score of 0.845.
Download
Paper Nr: 98
Title:

A Delay-Aware DRL-Based Environment for Cooperative Multi-UAV Systems in Multi-Purpose Scenarios
Authors:

Damiano Brunori and Luca Iocchi
Abstract: We provide a customizable environment based on Deep Reinforcement Learning (DRL) strategies for handling cooperative multi-UAV (Unmanned Aerial Vehicles) scenarios when delays are involved in the decision-making process for tasks such as spotting, tracking, coverage and many others. Users can choose among various combinations of tasks and parameters and customize the scenarios by implementing new desired functionalities. This environment provides the opportunity to compare different approaches, taking into account either implicitly or explicitly the delays applied to actions and observations. The awareness of the delay, along with the possible usage of real-world-based external files, increases the reality level of the environment by possibly easing the knowledge transferability process of the learned policy from the simulated environment to the real one. Finally, we show that use cases could generate new benchmarking tools for collaborative multi-UAV scenarios where DRL solutions must consider delays.
Download
Paper Nr: 101
Title:

Benchmarking Quantum Surrogate Models on Scarce and Noisy Data
Authors:

Jonas Stein, Michael Poppel, Philip Adamczyk, Ramona Fabry, Zixin Wu, Michael Kölle, Jonas Nüßlein, Daniëlle Schuman, Philipp Altmann, Thomas Ehmer, Vijay Narasimhan and Claudia Linnhoff-Popien
Abstract: Surrogate models are ubiquitously used in industry and academia to efficiently approximate black box functions. As state-of-the-art methods from classical machine learning frequently struggle to solve this problem accurately for the often scarce and noisy data sets in practical applications, investigating novel approaches is of great interest. Motivated by recent theoretical results indicating that quantum neural networks (QNNs) have the potential to outperform their classical analogs in the presence of scarce and noisy data, we benchmark their qualitative performance for this scenario empirically. Our contribution displays the first application-centered approach of using QNNs as surrogate models on higher dimensional, real world data. When compared to a classical artificial neural network with a similar number of parameters, our QNN demonstrates significantly better results for noisy and scarce data, and thus motivates future work to explore this potential quantum advantage. Finally, we demonstrate the performance of current NISQ hardware experimentally and estimate the gate fidelities necessary to replicate our simulation results.
Download
Paper Nr: 102
Title:

GREED: Graph Learning Based Relation Extraction with Entity and Dependency Relations
Authors:

Mohamed Y. Landolsi, Lobna Hlaoua and Lotfi Ben Romdhane
Abstract: A large number of electronic medical documents are generated by specialists, containing valuable information for various medical tasks such as medical prescriptions. Extracting this information from extensive natural language text can be challenging. Named Entity Recognition (NER) and Relation Extraction (RE) are key tasks in clinical information extraction. Systems often rely on machine learning and rule-based techniques. Modern methods involve dependency parsing and graph-based deep learning algorithms. However, the effectiveness of these techniques and certain features is not thoroughly studied. Additionally, it would be advantageous to properly integrate rules with deep learning models. In this paper, we introduce GREED (Graph learning based Relation Extraction with Entity and Dependency relations). GREED is based on graph classification using Graph Convolutional Network (GCN). We transform each sentence into a weighted graph via dependency parsing. Words are represented with features that capture co-occurrence, dependency type, entities, and relation verbs, with focus on the entity pair. Experiments on clinical records (i2b2/VA 2010) show that relevant features efficiently integrated with GCN achieve higher performance.
Download
Paper Nr: 107
Title:

New Speed Limit Recognition System: Software and Hardware Validation
Authors:

Nesrine Triki, Mohamed Karray and Mohamed Ksantini
Abstract: Recent advancements in intelligent driving have led to the integration of various automated systems into vehicles, including Speed Limit Recognition systems, which play a crucial role in enhancing road safety and autonomous driving technologies. This paper presents a comprehensive approach to Speed Limit Recognition, based on three modules: detection, classification, and the fusion of machine learning and deep learning classifiers. The proposed approach achieves impressive results, with an accuracy of 99.98% using Dempster Shafer theory and 99.96% with the voting technique. The system’s performance is rigorously evaluated through simulation and hardware validation using a Raspberry Pi 4 board. Experimental results indicate high performance rates across nine classes from the German Traffic sign Recognition Benchmark dataset in an average processing time of 0.15 seconds.
Download
Paper Nr: 108
Title:

Evaluating Quantum Support Vector Regression Methods for Price Forecasting Applications
Authors:

Horst Stühler, Daniel Pranjić and Christian Tutschku
Abstract: Support vector machines are powerful and frequently used machine learning methods for classification and regression tasks, which rely on the construction of kernel matrices. While crucial for the performance of this machine learning approach, choosing the most suitable kernel is highly problem-dependent. The emergence of quantum computers and quantum machine learning techniques provides new possibilities for generating powerful quantum kernels. Within this work, we solve a real-world price forecasting problem using fidelity and projected quantum kernels, which are promising candidates for the utility of near-term quantum computing. In our analysis, we examine and validate the most auspicious quantum kernels from literature and compare their performance with an optimized classical kernel. Unlike previous work on quantum support vector machines, our dataset includes categorical features that need to be encoded as numerical features, which we realize by using the one-hot-encoding scheme. One-hot-encoding, however, increases the dimensionality of the dataset significantly, which collides with the current limitations of noisy intermediate scale quantum computers. To overcome these limitations, we use autoencoders to learn a low-dimensional representation of the feature space that still maintains the most important information of the original data. To examine the impact of autoencoding, we compare the results of the encoded date with the results of the original, unencoded dataset. We could demonstrate that quantum kernels are comparable to or even better than the classical support vector machine kernels regarding the mean absolute percentage error scores for both encoded and unencoded datasets.
Download
Paper Nr: 111
Title:

Harnessing LLM Conversations for Goal Model Generation from User Reviews
Authors:

Shuaicai Ren, Hiroyuki Nakagawa and Tatsuhiro Tsuchiya
Abstract: User reviews are a valuable resource for developers, as the reviews contain requests for new features and bug reports. By conducting the requirements analysis of user reviews, developers can gain timely insights for the application, which is crucial for continuously enhancing user satisfaction. The goal model is a commonly used model during requirements analysis. Utilizing reviews to generate goal models can assist developers in understanding user requirements comprehensively. However, given the vast number of reviews, manually collecting reviews and creating goal models is a significant challenge. A method for clustering user reviews and automatically generating goal models has been proposed. Nevertheless, the accuracy of the goal models generated by this method is limited. To address these limitations of the existing method and enhance precision of goal model generation, we propose a goal-generation process based on Large Language Models (LLMs). This process does not directly generate goal models from user reviews; instead, it treats goal model generation as a clustering problem, allowing for the visualization of the relationship between reviews and goals. Experiments demonstrate that compared to the existing method, our LLM-based goal model generation process enhance the precision of goal model generation.
Download
Paper Nr: 113
Title:

Variational Quantum Circuit Design for Quantum Reinforcement Learning on Continuous Environments
Authors:

Georg Kruse, Theodora-Augustina Drăgan, Robert Wille and Jeanette Miriam Lorenz
Abstract: Quantum Reinforcement Learning (QRL) emerged as a branch of reinforcement learning (RL) that uses quantum submodules in the architecture of the algorithm. One branch of QRL focuses on the replacement of neural networks (NN) by variational quantum circuits (VQC) as function approximators. Initial works have shown promising results on classical environments with discrete action spaces, but many of the proposed architectural design choices of the VQC lack a detailed investigation. Hence, in this work we investigate the impact of VQC design choices such as angle embedding, encoding block architecture and postprocessesing on the training capabilities of QRL agents. We show that VQC design greatly influences training performance and heuristically derive enhancements for the analyzed components. Additionally, we show how to design a QRL agent in order to solve classical environments with continuous action spaces and benchmark our agents against classical feed-forward NNs.
Download
Paper Nr: 118
Title:

Investigation of the Performance of Different Loss Function Types Within Deep Neural Anchor-Free Object Detectors
Authors:

Ala’a Alshubbak and Daniel Görges
Abstract: In this paper, an investigation of different IoU loss functions and a spatial attention mechanism within anchor-free object detectors is presented. Two anchor-free dense predictor models are studied: FASF and FCOS models. The models are tested on two different datasets: the benchmark COCO dataset and a small dataset called OPEDD. The results show that some loss functions and using the attention mechanism outperform their original counterparts for both the huge multi-class COCO dataset and the small unity-class dataset of OPEDD. The proposed structure is tested over different backbones: ResNet-50, ResNet-101, and ResNeXt-101. The accuracy of basic models trained over the coco dataset improves by 1.3% and 1.6% mAP for the FSAF and FCOS models based on ResNet-50, respectively. On the other hand, it increases by 2.3% and 15.8% for the same models when trained on the OPEDD dataset. The effect is interpreted using a saliency map.
Download
Paper Nr: 121
Title:

Neural Bradley-Terry Rating: Quantifying Properties from Comparisons
Authors:

Satoru Fujii
Abstract: Many properties in the real world doesn’t have metrics and can’t be numerically observed, making them difficult to learn. To deal with this challenging problem, prior works have primarily focused on estimating those properties by using graded human scores as the target label in the training. Meanwhile, rating algorithms based on the Bradley-Terry model are extensively studied to evaluate the competitiveness of players based on their match history. In this paper, we introduce the Neural Bradley-Terry Rating (NBTR), a novel machine learning framework designed to quantify and evaluate properties of unknown items. Our method seamlessly integrates the Bradley-Terry model into the neural network structure. Moreover, we generalize this architecture further to asymmetric environments with unfairness, a condition more commonly encountered in real-world settings. Through experimental analysis, we demonstrate that NBTR successfully learns to quantify and estimate desired properties.
Download
Paper Nr: 124
Title:

Probabilistic Model Checking of Stochastic Reinforcement Learning Policies
Authors:

Dennis Gross and Helge Spieker
Abstract: We introduce a method to verify stochastic reinforcement learning (RL) policies. This approach is compatible with any RL algorithm as long as the algorithm and its corresponding environment collectively adhere to the Markov property. In this setting, the future state of the environment should depend solely on its current state and the action executed, independent of any previous states or actions. Our method integrates a verification technique, referred to as model checking, with RL, leveraging a Markov decision process, a trained RL policy, and a probabilistic computation tree logic (PCTL) formula to build a formal model that can be subsequently verified via the model checker Storm. We demonstrate our method’s applicability across multiple benchmarks, comparing it to baseline methods called deterministic safety estimates and naive monolithic model checking. Our results show that our method is suited to verify stochastic RL policies.
Download
Paper Nr: 128
Title:

Generating Products Placement in Warehouse Using BLPSO and MIQCPs
Authors:

Ayaka Sugiura, Takahiro Suzuki, Koya Ihara, Takuto Sakuma and Shohei Kato
Abstract: Expansion of the e-commerce market due to the development of the Internet has increased in the volume of distribution, and the number of operations in distribution warehouses had also increased. Picking operation is one of the most important tasks, and companies are trying to make this task more efficient by introducing autonomous mobile robots (AMRs), which transports products manually picked to a depot. In this study, we propose a method to generate product assignments that make picking operations more efficient through a two-step optimization process. First, product assignments for utilizing AMRs are generated using particle swarm optimization. Next, in-shelf products layout is generated by mathematical optimization for the products group assigned to the shelves. In product placement optimization, one of the approximate solution methods of the metaheuristic, BLPSO, is fused with a class-based warehouse to obtain an optimal solution. In addition, the problem of in-shelf product layout is formulated in MIQCPs. The constraint expression is used to generate a layout that considers preventing picking mistakes and ensuring the safety of the picker. We have conducted placement optimization experiments using real-world logistic data and discuss the effectiveness of the proposed method.
Download
Paper Nr: 130
Title:

A Comprehensive Analysis of Parkinson’s Disease Detection Through Inertial Signal Processing
Authors:

Manuel Gil-Martín, Sergio Esteban-Romero, Fernando Fernández-Martínez and Rubén San-Segundo
Abstract: When developing deep learning systems for Parkinson’s Disease (PD) detection using inertial sensors, a comprehensive analysis of some key factors, including data distribution, signal processing domain, number of sensors, and analysis window size, is imperative to refine tremor detection methodologies. Leveraging the PD-BioStampRC21 dataset with accelerometer recordings, our state-of-the-art deep learning architecture extracts a PD biomarker. Applying Fast Fourier Transform (FFT) magnitude coefficients as a preprocessing step improves PD detection in Leave-One-Subject-Out Cross-Validation (LOSO CV), achieving 66.90% accuracy with a single sensor and 6.4-second windows, compared to 60.33% using raw samples. Integrating information from all five sensors boosts performance to 75.10%. Window size analysis shows that 3.2-second windows of FFT coefficients from all sensors outperform shorter or longer windows, with a window-level accuracy of 80.49% and a user-level accuracy of 93.55% in a LOSO scenario.
Download
Paper Nr: 133
Title:

LACNN: A Deep Learning Model for Persian Question Analysis
Authors:

Fatemeh E. Khaksefidi, Afsaneh Fatemi, Mohammad A. Nematbakhsh and Mahsa A. Kia
Abstract: Question-answering systems, characterized by their three core functions of question classification, information retrieval, and answer selection, necessitate refinement to optimize their precision in retrieving exact answers. Question classification is a fundamental task that predicts the expected answer to a question. However, agglutinative languages constrained the performance of question classification algorithms, especially with inadequate and limited resources languages, such as Persian. In this paper, we proposed a multi-layer Long-short-term memory (LSTM) Attention Convolutional Neural Network (CNN)(LACNN) classifier model that extracts information from a Persian language context. The model operates autonomously without the need for previous knowledge and external features. Also, the first Persian open-domain medical question dataset, UIMQC, is proposed. UIMQC is the translation of the GARD dataset from English. The questions within UIMQC are highly technical and complex, often related to rare diseases that require diagnosis by specialists. The results showed that the model outperformed baseline methods by 9% on the UTQC dataset and achieved 67.08% accuracy on the UIMQC dataset. Therefore, we suggest the LACNN model for other morphological analysis tasks in different low-resource languages, as in Question Answering systems it improves the performance for retrieving accurate answers to the users’ queries.
Paper Nr: 134
Title:

Oral Diseases Recognition Based on Photographic Images
Authors:

Mazin S. Mohammed, Salah Zrigui and Mounir Zrigui
Abstract: Recently, the automation diagnosis process of dental caries plays a critical role in medical applications. This paper presents a new dataset of photo-graphic images for six different types of oral diseases. The dataset is gathered and labelled by professional medical operators in the dentistry field. We use the collected dataset to train a binary classifier to determine whether the region of interests (ROI) needs detection or not inside the input image. Then, we train a detector to detect and localize the required ROI. Finally, we use the detected regions to train a CNN network by adopting transfer learning technique to classify various kinds of teeth diseases. With this model, we obtained an almost 93 % accuracy by modifying and re-training the pre-trained model VGG19.
Download
Paper Nr: 138
Title:

Decoupling the Backward Pass Using Abstracted Gradients
Authors:

Kyle Rogers, Hao Yu, Seong-Eun Cho, Nancy Fulda, Jordan Yorgason and Tyler J. Jarvis
Abstract: In this work we introduce a novel method for decoupling the backward pass of backpropagation using mathematical and biological abstractions to approximate the error gradient. Inspired by recent findings in neuroscience, our algorithm allows gradient information to skip groups of layers during the backward pass, such that weight updates at multiple depth levels can be calculated independently. We explore both gradient abstractions using the identity matrix as well as an abstraction that we derive mathematically for network regions that consist of piecewise-linear layers (including layers with ReLU and leaky ReLU activations). We validate the derived abstraction calculation method on a fully connected network with ReLU activations. We then test both the derived and identity methods on the transformer architecture and show the capabilities of each method on larger model architectures. We demonstrate empirically that a network trained using an appropriately chosen abstraction matrix can match the loss and test accuracy of an unmodified network, and we provide a roadmap for the application of this method toward depth-wise parallelized models and discuss the potential of network modularization by this method.
Download
Paper Nr: 141
Title:

Flattening Based Cuckoo Search Optimization Algorithm for Community Detection in Multiplex Networks
Authors:

Randa Boukabene, Fatima Benbouzid-Si Tayeb and Narimene Dakiche
Abstract: Complex network analysis is a thriving research field, with a particular focus on community detection. This paper addresses the challenge of community detection in multiplex networks, which model multiple types of relationships to reflect reality. Our approach consists of two key steps. First, we employ multiplex network flattening techniques to transform it into a one-dimensional network. Second, we introduce a cuckoo search-based algorithm to maximize the modularity function and identify the best network partitions. Our algorithm strategically combines the continuous aspects of the standard cuckoo search algorithm with the discrete nature of community detection, to achieve better results. Experiments on both synthetic and real-world multiplex networks demonstrate the efficiency and effectiveness of our approach.
Download
Paper Nr: 143
Title:

Advanced Deep Learning Techniques for Industry 4.0: Application to Mechanical Design and Structural Health Monitoring
Authors:

Fakhreddine Ababsa
Abstract: Nowadays, Deep Learning (DL) techniques are increasingly employed in industrial applications. This paper investigate the development of data-driven models for two use cases: Additive Manufacturing-driven Topology Optimization and Structural Health Monitoring (SHM). We first propose an original data-driven generative method that integrates the mechanical and geometrical constraints concurrently at the same conceptual level and generates a 2D design accordingly. In this way, it adapts the geometry of the design to the manufacturing criteria, allowing the designer better interpretation and avoiding being stuck in a time-consuming loop of drawing the CAD and testing its performance. On the other hand, SHM technique is dedicated to the continuous and non-invasive monitoring of structures integrity, ensuring safety and optimal performances through on-site real-time measurements. We propose in this work new ways of structuring data that increase the accuracy of data driven SHM algorithms and that are based on the physical knowledge related with the structure to be inspected. We focus our study on the damage classification step within the aeronautic context, where the primary objective is to distinguish between different damage types in composite plates. Experimental results are presented to demonstrate the effectiveness of the proposed approaches.
Download
Paper Nr: 149
Title:

A Novel Image Steganography Method Based on Spatial Domain with War Strategy Optimization and Reed Solomon Model
Authors:

Hassan J. Azooz, Khawla Ben Salah, Monji Kherallah and Mohamed S. Naceur
Abstract: In this paper, we propose a novel approach to steganography using the War Search Optimization (WSO) algorithm. Steganography is the practice of concealing messages within other data, such as images or audio files. Our approach employs the WSO algorithm to optimize the parameters of a steganography algorithm, aiming to maximize the perceptual similarity between the cover image and the stego image. We demonstrate the effectiveness of our approach on a variety of cover images and secret messages and show that our method produces stego images with high perceptual similarity to the cover images. Our results suggest that the WSO algorithm is a promising tool for optimizing steganography algorithms. Also, this paper presents a new approach to steganography that utilizes the War Search Optimization (WSO) algorithm. Steganography involves hiding messages within other data, such as images or audio files. Our method applies the WSO algorithm to optimize the parameters of a steganography algorithm with the goal of maximizing the perceptual similarity between the cover image and the stego image. We evaluate our approach on various cover images and secret messages and demonstrate that our technique generates stego images with high perceptual similarity to the cover images. The results indicate that the WSO algorithm is a valuable tool for optimizing steganography algorithms.
Download
Paper Nr: 156
Title:

From Targets to Rewards: Continuous Target Sets in the Algorithmic Search Framework
Authors:

Milo Knell, Sahil Rane, Forrest Bicker, Tiger Che, Alan Wu and George D. Montañez
Abstract: Many machine learning tasks have a measure of success that is naturally continuous, such as error under a loss function. We generalize the Algorithmic Search Framework (ASF), used for modeling machine learning domains as discrete search problems, to the continuous space. Moving from discrete target sets to a continuous measure of success extends the applicability of the ASF by allowing us to model fundamentally continuous notions like fuzzy membership. We generalize many results from the discrete ASF to the continuous space and prove novel results for a continuous measure of success. Additionally, we derive an upper bound for the expected performance of a search algorithm under arbitrary levels of quantization in the success measure, demonstrating a negative relationship between quantization and the performance upper bound. These results improve the fidelity of the ASF as a framework for modeling a range of machine learning and artificial intelligence tasks.
Download
Paper Nr: 161
Title:

System-Call-Level Dynamic Analysis for Code Translation Candidate Selection
Authors:

Narumi Yoneda, Ryo Hatano and Hiroyuki Nishiyama
Abstract: In this study, we propose a methodology that uses dynamic analysis (DA) data to select better code-translation candidates. For the DA data, we recorded the history of system-call invocations to understand the actions of the program during execution, providing insights independent of the programming language. We implemented and publicized a DA system, which enabled a fully automated analysis. In our method, we generated multiple translation candidates for programming languages using TransCoder. Subsequently, we performed DA on all the generated candidates and original code. For optimal selection, we compared the DA data of the original code with the generated data and calculated the similarity. To compare the DA data, we used natural language processing techniques on DA data to fix the sequence length. We also attempted to directly compare the variable-length system-call sequences. In this study, we demonstrated that the characteristics of system-call invocations vary even within the same code. For instance, the order of invocations and the number of times the same system-calls an invocation differ. We discuss the elimination of these uncertainties when comparing system-calls.
Download
Paper Nr: 164
Title:

Dynamic Path Planning for Autonomous Vehicles: A Neuro-Symbolic Approach
Authors:

Omar Elrasas, Nourhan Ehab, Yasmin Mansy and Amr El Mougy
Abstract: The rise of autonomous vehicles has transformed transportation, promising safer and more efficient mobility. Dynamic path planning is crucial in autonomous driving, requiring real-time decisions for navigating complex environments. Traditional approaches, like rule-based methods or pure machine learning, have limitations in addressing these challenges. This paper explores integrating Neuro-Symbolic Artificial Intelligence (AI) for dynamic path planning in self-driving cars, creating two regression models with the Logic Tensor Networks (LTN) Neuro-Symbolic framework. Tested on the CARLA simulator, the project effectively followed road lanes, avoided obstacles, and adhered to speed limits. Root mean square deviation (RMSE) gauged the LTN models’ performance, revealing significant improvement, particularly with small datasets, showcasing Neuro-Symbolic AI’s data efficiency. However, LTN models had longer training times compared to linear and XGBoost regression models.
Download
Paper Nr: 167
Title:

Explainability Insights to Cellular Simultaneous Recurrent Neural Networks for Classical Planning
Authors:

Michaela Urbanovská and Antonín Komenda
Abstract: The connection between symbolic artificial intelligence and statistical machine learning has been explored in many ways. That includes using machine learning to learn new heuristic functions for navigating classical planning algorithms. Many approaches which target this task use different problem representations and different machine learning techniques to train estimators for navigating search algorithms to find sequential solutions to deterministic problems. In this work, we focus on one of these approaches which is the semantically layered Cellular Simultaneous Neural Network architecture (slCSRN) (Urbanovská and Komenda, 2023) used to learn heuristic for grid-based planning problems represented by the semantically layered representation. We create new problem domains for this architecture - the Tetris and Rush-Hour domains. Both do not have an explicit agent that only modifies its surroundings unlike already explored problem domains. We compare the performance of the trained slCSRN to the existing classical planning heuristics and we also provide insights into the slCSRN computation as we provide explainability analysis of the learned heuristic functions.
Download
Paper Nr: 171
Title:

Knowledge-Aware Object Detection in Traffic Scenes
Authors:

Jean-Francois Nies, Syed Tahseen Raza Rizvi, Mohsin Munir, Ludger V. Elst and Andreas Dengel
Abstract: Autonomous driving is a widely popular domain that empowers the autonomous vehicle to make crucial decisions in a constantly evolving traffic scenario. The role of perception is pivotal in the secure operation of the autonomous vehicle in a complex traffic scene. Recently, several approaches have been proposed for the task of object detection. In this paper, we demonstrate that the concept of Semantic Consistency and the ensuing method of Knowledge-Aware Re-Optimization can be adapted for the problem of object detection in intricate traffic scenes. Moreover, we also introduce a novel method for extracting a knowledge graph encoding the semantic relationship between the traffic participants from an autonomous driving dataset. We also conducted an investigation into the efficacy of utilizing diverse knowledge graph generation methodologies and in- and out-domain knowledge sources on the efficacy of the outcomes. Finally, we investigated the effectiveness of knowledge-aware re-optimization on the Faster-RCNN and DETR object detection models. Results suggest that modest but consistent improvements in precision and recall can be achieved using this method.
Download
Paper Nr: 176
Title:

Adaptive Questionnaire Design Using AI Agents for People Profiling
Authors:

Ciprian Paduraru, Rares Cristea and Alin Stefanescu
Abstract: Creating employee questionnaires, surveys or evaluation forms for people to understand various aspects such as motivation, improvement opportunities, satisfaction, or even potential cybersecurity risks is a common practice within organizations. These surveys are usually not tailored to the individual and have a set of predetermined questions and answers. The objective of this paper is to design AI agents that are flexible and adaptable in choosing the survey content for each individual according to their personality. The developed framework is open source, generic and can be adapted to many use cases. For the evaluation, we present a real-world use case of detecting potentially inappropriate behavior in the workplace. The results obtained are promising and suggest that the decision algorithms for content selection approaches and personalized surveys via AI agents are similar to a real human resource manager in our use case.
Download
Paper Nr: 178
Title:

Designing Algorithms for the Shortest Path Reconfiguration Problem Using Decision Diagram Operations
Authors:

Shou Ooba, Jun Kawahara and Shin-ichi Minato
Abstract: This paper proposes decision diagram (DD)-based algorithms for the (edge-unweighted) shortest s-t path reconfiguration problem. In the problem, given a graph and two shortest s-t paths, the task is to decide whether one shortest path can be transformed into the other one by repeatedly applying the reconfiguration rule to the path, where the reconfiguration rule is to change one vertex of the path at a time while maintaining shortest s-t paths. We propose several DD-based algorithms for the problem and confirm their performance by computer experiments. We succeeded in finding a shortest reconfiguration sequence with length 961,012 in 629.0 seconds for some instance.
Download
Paper Nr: 182
Title:

Disentangling Quantum and Classical Contributions in Hybrid Quantum Machine Learning Architectures
Authors:

Michael Kölle, Jonas Maurer, Philipp Altmann, Leo Sünkel, Jonas Stein and Claudia Linnhoff-Popien
Abstract: Quantum computing offers the potential for superior computational capabilities, particularly for data-intensive tasks. However, the current state of quantum hardware puts heavy restrictions on input size. To address this, hybrid transfer learning solutions have been developed, merging pre-trained classical models, capable of handling extensive inputs, with variational quantum circuits. Yet, it remains unclear how much each component – classical and quantum – contributes to the model’s results. We propose a novel hybrid architecture: instead of utilizing a pre-trained network for compression, we employ an autoencoder to derive a compressed version of the input data. This compressed data is then channeled through the encoder part of the autoencoder to the quantum component. We assess our model’s classification capabilities against two state-of-the-art hybrid transfer learning architectures, two purely classical architectures and one quantum architecture. Their accuracy is compared across four datasets: Banknote Authentication, Breast Cancer Wisconsin, MNIST digits, and AudioMNIST. Our research suggests that classical components significantly influence classification in hybrid transfer learning, a contribution often mistakenly ascribed to the quantum element. The performance of our model aligns with that of a variational quantum circuit using amplitude embedding, positioning it as a feasible alternative.
Download
Paper Nr: 183
Title:

Using Ensemble Models for Malicious Web Links Detection
Authors:

Claudia-Ioana Coste
Abstract: Web technology advances faster than humans can adapt to it and develop the proper online skills. Most users are not experienced enough to have a good online knowledge on how to protect their data. Thus, many people can become vulnerable to threats. The most common online attacks are through malicious web links, which can deceive users into clicking them and running malicious code. The present approach proposed to advance the field of malicious web links detection through ensemble models by developing a nature-inspired ensemble. Our methodology is tested against two datasets, and we conduct an additional calibration step for all the models. For the first database, we managed to improve the detection accuracy from other solutions, by achieving 97.05%. In the case of the second dataset, our empirical strategy is not accurate enough, reaching just 91.12% accuracy. The proposed ensemble is heterogeneous, having a weight voting mechanism, where weights are generated with the Particle Swarm Optimization algorithm. To build the ensemble we compared 12 individual machine learning models, including Logistic Regression, Support Vector Machine, Adaptive Boosting, Random Forest, Decision Tree, K-Nearest Neighbor, Perceptron, Nearest Centroid, Passive Aggressive Classifier, Stochastic Gradient Descent, KMeans, and different variants for Naive Bayes.
Download
Paper Nr: 191
Title:

A Unique Training Strategy to Enhance Language Models Capabilities for Health Mention Detection from Social Media Content
Authors:

Pervaiz I. Khan, Muhammad N. Asim, Andreas Dengel and Sheraz Ahmed
Abstract: An ever-increasing amount of social media content requires advanced AI-based computer programs capable of extracting useful information. Specifically, the extraction of health-related content from social media is useful for the development of diverse types of applications including disease spread, mortality rate prediction, and finding the impact of diverse types of drugs on diverse types of diseases. Language models are competent in extracting the syntactic and semantics of text. However, they face a hard time extracting similar patterns from social media texts. The primary reason for this shortfall lies in the non-standardized writing style commonly employed by social media users. Following the need for an optimal language model competent in extracting useful patterns from social media text, the key goal of this paper is to train language models in such a way that they learn to derive generalized patterns. The key goal is achieved through the incorporation of random weighted perturbation and contrastive learning strategies. On top of a unique training strategy, a meta predictor is proposed that reaps the benefits of 5 different language models for discriminating posts of social media text into non-health and health-related classes. Comprehensive experimentation across 3 public benchmark datasets reveals that the proposed training strategy improves the performance of the language models up to 3.87%, in terms of F1-score, as compared to their performance with traditional training. Furthermore, the proposed meta predictor outperforms existing health mention classification predictors across all 3 benchmark datasets.
Download
Paper Nr: 199
Title:

ConEX: A Context-Aware Framework for Enhancing Explanation Systems
Authors:

Yasmeen Khaled and Nourhan Ehab
Abstract: Recent advances in Artificial Intelligence (AI) have led to the widespread adoption of intricate AI models, raising concerns about their opaque decision-making. Explainable AI (XAI) is crucial for improving transparency and trust. However, current XAI approaches often prioritize AI experts, neglecting broader stakeholder requirements. This paper introduces a comprehensive context taxonomy and ConEX, an adaptable framework for context-sensitive explanations. ConEX includes explicit problem-solving knowledge and contextual insights, allowing tailored explanations for specific contexts. We apply the framework to personalize movie recommendations by aligning explanations with user profiles. Additionally, we present an empirical user study highlighting diverse preferences for contextualization depth in explanations, highlighting the importance of catering to these preferences to foster trust and satisfaction in AI systems.
Download
Paper Nr: 201
Title:

Action Duration Generalization for Exact Multi-Agent Collective Construction
Authors:

Martin Rameš and Pavel Surynek
Abstract: This paper addresses exact approaches to multi-agent collective construction problem which tasks a group of cooperative agents to build a given structure in a blocksworld under the gravity constraint. We propose a generalization of the existing exact model based on mixed integer linear programming by accommodating varying agent action durations. We refer to the model as a fraction-time model. The introduction of action durations enables one to create a more realistic model for various domains. It provides a significant reduction of plan execution duration at the cost of increased computational time, which rises steeply the closer the model gets to the exact real-world action duration. We also propose a makespan estimation function for the fraction-time model. This can be used to estimate the construction time reduction size for cost-benefit analysis. The fraction-time model and the makespan estimation function have been evaluated in a series of experiments using a set of benchmark structures. The results show a significant reduction of plan execution duration for non-constant duration actions due to decreasing synchronization overhead at the end of each action. According to the results, the makespan estimation function provides a reasonably accurate estimate of the makespan.
Download
Paper Nr: 202
Title:

Facial Expression-Based Drowsiness Detection System for Driver Safety Using Deep Learning Techniques
Authors:

Amina Turki, Sirine Ammar, Mohamed Karray and Mohamed Ksantini
Abstract: Driver drowsiness is a leading cause of road accidents, resulting in severe physical injuries, fatalities, and substantial economic losses. To address this issue, a sophisticated Driver Drowsiness Detection (DDD) system is needed to alert the driver in case of abnormal behaviour and prevent potential catastrophes. The proposed DDD system calculates the Eyes Closure Ratio (ECR) and Mouth Opening Ratio (MOR) using the Chebyshev distance, instead of the classical Euclidean distance, to model the driver’s behaviour and to detect drowsiness states. This system uses simple camera and deep transfer learning techniques to detect the driver’s drowsiness state and then alert the driver in real time situations. The system achieves 96% for the VGG19 model, and 98% for the ResNet50 model, with a precision rate of 98% in assessing the driver’s dynamics.
Download
Paper Nr: 208
Title:

Inferring Interpretable Semantic Cognitive Maps from Noisy Document Corpora
Authors:

Yahya Emara, Tristan Weger, Ryan Rubadue, Rishabh Choudhary, Simona Doboli and Ali A. Minai
Abstract: With the emergence of deep learning-based semantic embedding models, it has become possible to extract large-scale semantic spaces from text corpora. Semantic elements such as words, sentences and documents can be represented as embedding vectors in these spaces, allowing their use in many applications. However, these semantic spaces are very high-dimensional and the embedding vectors are hard to interpret for humans. In this paper, we demonstrate a method for obtaining more meaningful, lower-dimensional semantic spaces, or cognitive maps, through the semantic clustering of the high-dimensional embedding vectors obtained from a real-world corpus. A key limitation in this is the presence of semantic noise in real-world document corpora. We show that pre-filtering the documents for semantic relevance can alleviate this problem, and lead to highly interpretable cognitive maps.
Download
Paper Nr: 211
Title:

A Machine Learning Approach Using Interpretable Models for Predicting Success of NCAA Basketball Players to Reach NBA
Authors:

Dante A. Costa, Joseana M. Fechine, José S. Brito, João R. Ferro, Evandro B. Costa and Roberta V. Lopes
Abstract: Predictive models in machine learning and knowledge discovery in databases have been used in various application domains, including sports and basketball, in the context of the National Basketball Association (NBA), where one can find relevant predictive issues. In this paper, we apply supervised machine learning to examine historical and statistical data and features from players in the NCAA basketball league, addressing the prediction problem of automatically identifying NCAA basketball players with an excellent chance of reaching the NBA and becoming successful. This problem is not easy to resolve; among other difficulties, many factors and high uncertainty can influence basketball players’ success in the mentioned context. One of our main motivations for addressing this predicting problem is to provide decision-makers with relevant information, helping them to improve their hiring judgment. To this end, we aim to have the advantage of producing an interpretable prediction model representation and satisfactory accuracy levels, therefore, considering a trade-off between Interpretability and Predictive Accuracy, we have invested in white-box classification methods, such as induction of decision trees, as well as logistic regression. However, as a baseline, we have considered a relevant method as a reference for the black-box model. Furthermore, in our approach, we explored these methods combined with genetic algorithms to improve their predictive accuracy and promote feature reduction. The results have been thoroughly compared, and models exhibiting superior performance have been emphasized, revealing predictive accuracy differences between the best white box and black box models were very small. The pairing of the genetic algorithm and logistic regression was particularly noteworthy, outperforming other models’ predictive accuracy and significant feature reduction, assisting the interpretability of the results. Furthermore, the analysis also highlighted which features were most important in the model.
Download
Paper Nr: 212
Title:

Mobile Agents-Based Framework for Dynamic Resource Allocation in Cloud Computing
Authors:

Safia Rabaaoui, Héla Hachicha and Ezzeddine Zagrouba
Abstract: Nowadays, cloud computing is becoming the more popular technology for various companies and consumers, who benefit from its increased efficiency, cost optimization, data security, unlimited storage capacity, etc. One of the biggest challenges of cloud computing is resource allocation. Its efficiency directly influences the performance of the whole cloud environment. Finding an effective method to address these critical issues and increase cloud performance was necessary. This paper proposes a mobile agents-based framework for dynamic resource allocation in cloud computing to minimize the cost of virtual machines and the makespan. Furthermore, its impact on the best response time and task rejection rate has been studied. The simulation shows that our method gave better results than the former ones.
Download
Paper Nr: 215
Title:

Out of the Cage: How Stochastic Parrots Win in Cyber Security Environments
Authors:

Maria Rigaki, Ondřej Lukáš, Carlos Catania and Sebastian Garcia
Abstract: Large Language Models (LLMs) have gained widespread popularity across diverse domains involving text generation, summarization, and various natural language processing tasks. Despite their inherent limitations, LLM-based designs have shown promising capabilities in planning and navigating open-world scenarios. This paper introduces a novel application of pre-trained LLMs as agents within cybersecurity network environments, focusing on their utility for sequential decision-making processes. We present an approach wherein pre-trained LLMs are leveraged as attacking agents in two reinforcement learning environments. Our proposed agents demonstrate similar or better performance against state-of-the-art agents trained for thousands of episodes in most scenarios and configurations. In addition, the best LLM agents perform similarly to human testers of the environment without any additional training process. This design highlights the potential of LLMs to address complex decision-making tasks within cybersecurity efficiently. Furthermore, we introduce a new network security environment named NetSecGame. The environment is designed to support complex multi-agent scenarios within the network security domain eventually. The proposed environment mimics real network attacks and is designed to be highly modular and adaptable for various scenarios.
Download
Paper Nr: 217
Title:

Explainability-Driven Leaf Disease Classification Using Adversarial Training and Knowledge Distillation
Authors:

Sebastian-Vasile Echim, Iulian-Marius Tăiatu, Dumitru-Clementin Cercel and Florin Pop
Abstract: This work focuses on plant leaf disease classification and explores three crucial aspects: adversarial training, model explainability, and model compression. The models’ robustness against adversarial attacks is enhanced through adversarial training, ensuring accurate classification even in the presence of threats. Leveraging explainability techniques, we gain insights into the model’s decision-making process, improving trust and transparency. Additionally, we explore model compression techniques to optimize computational efficiency while maintaining classification performance. Through our experiments, we determine that on a benchmark dataset, the robustness can be the price of the classification accuracy with performance reductions of 3%-20% for regular tests and gains of 50%-70% for adversarial attack tests. We also demonstrate that a student model can be 15-25 times more computationally efficient for a slight performance reduction, distilling the knowledge of more complex models.
Download
Paper Nr: 218
Title:

Compressing UNSAT CDCL Trees with Caching
Authors:

Anthony Blomme, Daniel L. Berre, Anne Parrain and Olivier Roussel
Abstract: We aim at providing users of SAT solvers with small, easily understandable proofs of unsatisfiability. Caching techniques have been proposed to identify redundant subproofs and reduce the size of some UNSAT proof trees. Branches are pruned when they correspond to subformulas that were proved unsatisfiable earlier in the tree. A caching mechanism based on subgraph isomorphism was proposed as postprocessing step both in the DPLL and CDCL architectures but the technique could only be integrated during the search on the DPLL architecture. This paper presents how to integrate such caching mechanism during the search for the CDCL case and presents a generalized caching mechanism for that architecture.
Download
Paper Nr: 223
Title:

Evaluating Large Language Models in Semantic Parsing for Conversational Question Answering over Knowledge Graphs
Authors:

Phillip Schneider, Manuel Klettner, Kristiina Jokinen, Elena Simperl and Florian Matthes
Abstract: Conversational question answering systems often rely on semantic parsing to enable interactive information retrieval, which involves the generation of structured database queries from a natural language input. For information-seeking conversations about facts stored within a knowledge graph, dialogue utterances are transformed into graph queries in a process that is called knowledge-based conversational question answering. This paper evaluates the performance of large language models that have not been explicitly pre-trained on this task. Through a series of experiments on an extensive benchmark dataset, we compare models of varying sizes with different prompting techniques and identify common issue types in the generated output. Our results demonstrate that large language models are capable of generating graph queries from dialogues, with significant improvements achievable through few-shot prompting and fine-tuning techniques, especially for smaller models that exhibit lower zero-shot performance.
Download
Paper Nr: 224
Title:

Fair and Equitable Machine Learning Algorithms in Healthcare: A Systematic Mapping
Authors:

Marcelo S. Mattos, Sean M. Siqueira and Ana B. Garcia
Abstract: Artificial intelligence (AI) is being employed in many fields, including healthcare. While AI has the potential to improve people’s lives, it also raises ethical questions about fairness and bias. This article reviews the challenges and proposed solutions for promoting fairness in medical decisions aided by AI algorithms. A systematic mapping study was conducted, analyzing 37 articles on fairness in machine learning in healthcare from five sources: ACM Digital Library, IEEE Xplore, PubMed, ScienceDirect, and Scopus. The analysis reveals a growing interest in the field, with many recent publications. The study offers an up-to-date and comprehensive overview of approaches and limitations for evaluating and mitigating biases, unfairness, and discrimination in healthcare-focused machine learning algorithms. This study’s findings provide valuable insights for developing fairer, equitable, and more ethical AI systems for healthcare.
Download
Paper Nr: 227
Title:

Autonomous Methods in Multisensor Architecture for Smart Surveillance
Authors:

Dani Manjah, Stéphane Galland, Christophe De Vleeschouwer and Benoît Macq
Abstract: This paper considers the deployment of flexible and high-performance surveillance systems. These systems must continuously integrate new sensors and sensing algorithms, which are autonomous (e.g., capable of making decisions independently of a central system) and possess interaction skills (e.g., capable of exchanging observations). For this purpose, our work proposes adopting an agent-based architecture derived from an organizational and holonic (i.e., system of systems) multi-agent model. It leverages autonomous processing methods, resulting in a scalable and modular multisensor and multimethod surveillance systems. A vehicle tracking case study demonstrates the relevance of our approach in terms of effectiveness and runtime.
Download
Paper Nr: 230
Title:

Medi-CAT: Contrastive Adversarial Training for Medical Image Classification
Authors:

Pervaiz I. Khan, Andreas Dengel and Sheraz Ahmed
Abstract: There are not many large medical image datasets available. Too small deep learning models can’t learn useful features, so they don’t work well due to underfitting, and too big models tend to overfit the limited data. As a result, there is a compromise between the two issues. This paper proposes a training strategy to overcome the aforementioned issues in medical imaging domain. Specifically, it employs a large pre-trained vision transformers to overcome underfitting and adversarial and contrastive learning techniques to prevent overfitting. The presented method has been trained and evaluated on four medical image classification datasets from the MedMNIST collection. Experimental results indicate the effectiveness of the method by improving the accuracy up-to 2% on three benchmark datasets compared to well-known approaches and up-to 4.1% over the baseline methods. Code can be accessed at: https://github.com/pervaizniazi/medicat.
Download
Paper Nr: 237
Title:

Class Anchor Margin Loss for Content-Based Image Retrieval
Authors:

Alexandru Ghita and Radu T. Ionescu
Abstract: Loss functions play a major role in influencing the effectiveness of neural networks in content-based image retrieval (CBIR). Existing loss functions can be categorized into metric learning and statistical learning. Metric learning often lacks efficiency due to pair mining, while statistical learning does not yield compact features. To this end, we introduce a novel repeller-attractor loss based on metric learning, which directly optimizes the L2 metric, without pair generation. Our novel loss comprises three terms: one to ensure features are attracted to class anchors, one that enforces anchor separability, and one that prevents anchor collapse. We evaluate our objective, applied to both convolutional and transformer architectures, on CIFAR-100, Food-101, SVHN, and ImageNet-200, showing that it outperforms existing functions in CBIR.
Download
Paper Nr: 239
Title:

A Toolset for Constraint Programming
Authors:

Thibault Falque and Romain Wallon
Abstract: Constraint Programming (CP) allows solving combinatorial problems across various domains. Numerous solvers and tools have been developed in this area. However, their interoperability is often limited. This paper presents a suite of tools for constraint programming, consisting of a solver interface and a remote control application. The solver interface offers a unified API for interacting with different solvers of various programming languages. Based on this API, we present a remote control system enabling to configure the solver and to observe and analyze its behaviour while it is running.
Download
Paper Nr: 245
Title:

A Description Language for Similarity, Belief Change and Trust
Authors:

Aaron Hunter
Abstract: We introduce a simple framework for describing and reasoning about situations where an agent receives information reported from external sources, and these reports cause them to change their beliefs. Our framework is inspired by classic action description languages, which use sets of causal statments to specify action effects in terms of transition systems. We suggest that this style of language can effectively capture important properties of similarity and trust, which are required to perform belief revision in practical settings. The language introduced in this paper allows us to specify a similarity relation on states, and it also allows us to explicitly associate an incoming report with a specific formula to be used as the input for a suitable belief revision operator. The result is a flexible framework that can describe a variety of belief change functions, and it can also capture the trust that is held in the reporting agent in a simple and transparent way. We demonstrate the connection with existing trust-influenced models of belief change. We then consider a speculative application where we apply our framework to reason about the correctness of trusted third party protocols. Directions for future work are considered.
Download
Paper Nr: 251
Title:

A Novel Partitioning Approach for Real-Time Scheduling of Mixed-Criticality Systems
Authors:

Hayfa Ben Abdallah, Hamza Gharsellaoui and Sadok Bouamama
Abstract: In real-time system (RTSys), a program is split into small tasks and distributed among several computing elements to minimize the overall system cost. Intrinsically, tasks allocation problem is NP- hard. To overcome this issue, it is necessary to introduce heuristics for generating near optimal solution to the given problem. This paper deals with the problem of dependent and periodic tasks to be assigned to different cores interconnected by a network-on-chip (NoC) in such a way that the load on each Core is almost acceptable. Further, the development of an effective algorithm for allocating ‘N’ tasks to ‘P’ cores. The system using task clustering to reduce the Communication Cost on the NoC. Experiment results and simulations demonstrate the efficiency of the proposed approach.
Download
Paper Nr: 252
Title:

SynCRF: Syntax-Based Conditional Random Field for TRIZ Parameter Minings
Authors:

Guillaume Guarino, Ahmed Samet and Denis Cavallucci
Abstract: Conditional random fields (CRF) are widely used for sequence labeling such as Named Entity Recognition (NER) problems. Most CRFs, in Natural Language Processing (NLP) tasks, model the dependencies between predicted labels without any consideration for the syntactic specificity of the document. Unfortunately, these approaches are not flexible enough to consider grammatically rich documents like patents. Additionally, the position and the grammatical class of the words may influence the text’s understanding. Therefore, in this paper, we introduce SynCRF which considers grammatical information to compute pairwise potentials. Syn-CRF is applied to TRIZ (Theory of Inventive Problem Solving), which offers a comprehensive set of tools to analyze and solve problems. TRIZ aims to provide users with inventive solutions given technical contradiction parameters. SynCRF is applied to mine these parameters from patent documents. Experiments on a labeled real-world dataset of patents show that SynCRF outperforms state-of-the-art and baseline approaches.
Download
Paper Nr: 258
Title:

Enhancing Manufacturing Quality Prediction Models Through the Integration of Explainability Methods
Authors:

Dennis Gross, Helge Spieker, Arnaud Gotlieb and Ricardo Knoblauch
Abstract: This research presents a method that utilizes explainability techniques to amplify the performance of machine learning (ML) models in forecasting the quality of milling processes, as demonstrated in this paper through a manufacturing use case. The methodology entails the initial training of ML models, followed by a fine-tuning phase where irrelevant features identified through explainability methods are eliminated. This procedural refinement results in performance enhancements, paving the way for potential reductions in manufacturing costs and a better understanding of the trained ML models. This study highlights the usefulness of explainability techniques in both explaining and optimizing predictive models in the manufacturing realm.
Download
Paper Nr: 259
Title:

Hybrid PSO-Based Rule Classifier for Disease Detection
Authors:

Cecilia Mariciuc and Madalina Raschip
Abstract: The application of data mining techniques in healthcare is common because the decision-making process for the diagnosis of medical conditions could benefit from the information extracted. A decision system must not only be accurate but also provide understandable explanations for its reasoning. Rule-based models seek to find a small set of rules that can effectively categorize data while providing great human readability. Rule discovery is a complex optimization problem, making it a good candidate for the application of PSO, a versatile, intuitive search algorithm. In this paper, a particle swarm optimization algorithm is used for learning classification rules as part of a Covering-based rule classifier. The proposed PSO is hybridized with the Iterated Local Search metaheuristic, and association rules are used as part of the initialization step. The classifier is tested on several unbalanced medical disease datasets with different types of attributes to more faithfully reflect real-world data. When compared with state-of-the-art rule-based classifiers, the studied algorithm shows good results and is highly interpretable.
Download
Paper Nr: 267
Title:

Quantum Federated Learning for Image Classification
Authors:

Leo Sünkel, Philipp Altmann, Michael Kölle and Thomas Gabor
Abstract: Federated learning is a technique in classical machine learning in which a global model is collectively trained by a number of independent clients, each with their own datasets. Using this learning method, clients are not required to reveal their dataset as it remains local; clients may only exchange parameters with each other. As the interest in quantum computing and especially quantum machine learning is steadily increasing, more concepts and approaches based on classical machine learning principles are being applied to the respective counterparts in the quantum domain. Thus, the idea behind federated learning has been transferred to the quantum realm in recent years. In this paper, we evaluate a straightforward approach to quantum federated learning using the widely used MNIST dataset. In this approach, we replace a classical neural network with a variational quantum circuit, i.e., the global model as well as the clients are trainable quantum circuits. We run three different experiments which differ in number of clients and data-subsets used. Our results demonstrate that basic principles of federated learning can be applied to the quantum domain while still achieving acceptable results. However, they also illustrate that further research is required for scenarios with increasing number of clients.
Download
Paper Nr: 268
Title:

AccidentGPT: Large Multi-Modal Foundation Model for Traffic Accident Analysis
Authors:

Kebin Wu, Wenbin Li and Xiaofei Xiao
Abstract: Traffic accident analysis is pivotal for enhancing public safety and developing road regulations. Traditional approaches, although widely used, are often constrained by manual analysis processes, subjective decisions, uni-modal outputs, as well as privacy issues related to sensitive data. This paper introduces the idea of AccidentGPT, a foundation model of traffic accident analysis, which incorporates multi-modal input data to automatically reconstruct the accident process video with dynamics details, and furthermore provide multi-task analysis with multi-modal outputs. The design of the AccidentGPT is empowered with a multi-modality prompt with feedback for task-oriented adaptability, a hybrid training schema to leverage labelled and unla-belled data, and a edge-cloud split configuration for data privacy. To fully realize the functionalities of this model, we proposes several research opportunities. This paper serves as the stepping stone to fill the gaps in traditional approaches of traffic accident analysis and attract the research community’s attention for automatic, objective, and privacy-preserving traffic accident analysis.
Download
Paper Nr: 274
Title:

RLAR: A Reinforcement Learning Abductive Reasoner
Authors:

Mostafa ElHayani
Abstract: Machine learning (ML) algorithms are the foundation of the modern AI environment. They are renowned for their capacity to solve complicated problems and generalize across a wide range of datasets. Nevertheless, a noteworthy disadvantage manifests itself as a lack of explainability. Symbolic AI is at the other extreme of the spectrum; in this case, every inference is a proof, allowing for transparency and traceability throughout the decision-making process. This paper proposes the Reinforcement Learning Abductive Reasoner (RLAR). A combination of modern and symbolic AI algorithms aimed to bridge the gap and utilize the best features of both methods. A case study has been chosen to test the implementation of the proposed reasoner. A knowledge-base (KB) vectorization step is implemented, and a Machine Learning model architecture is built to learn explanation inference. Furthermore, a simple abductive reasoner is also implemented to compare both approaches.
Download
Paper Nr: 278
Title:

SIFT-ResNet Synergy for Accurate Scene Word Detection in Complex Scenarios
Authors:

Riadh Harizi, Rim Walha and Fadoua Drira
Abstract: Scene text detection is of growing importance due to its various applications. Deep learning-based systems have proven effective in detecting horizontal text in natural scene images. However, they encounter difficulties when confronted with oriented and curved text. To tackle this issue, our study introduces a hybrid scene text detector that combines selective search with SIFT-based keypoint density analysis and a deep learning training architecture framework. More precisely, we investigated SIFT keypoints to identify important areas in an image for precise word localization. Then, we fine-tuned these areas with a deep learning-powered bounding box regressor. This combination ensured accurate word boundary alignment and enhancing word detection efficiency. We evaluated our method on benchmark datasets, including ICDAR2013, ICDAR2015, and SVT, comparing it with established state-of-the-art scene text detectors. The results underscore the strong performance of our scene text detector when dealing with complex scenarios.
Download
Paper Nr: 280
Title:

Digital Twin and Foundation Models: A New Frontier
Authors:

Athanasios Trantas and Paolo Pileggi
Abstract: A Foundation Model (FM) possesses extensive learning capabilities; it learns from diverse datasets. This is our opportunity to enhance the functionality of Digital Twin (DT) solutions in various sectors. The integration of FMs into the DT application is particularly relevant due to the increased prevalence of Artificial Intelligence (AI) in real-world applications. In this position paper, we begin to explain a novel perspective on this integration by exploring the potential of enhanced predictive analytics, adaptive learning, and improved handling of complex data within DTs — by way of designated purposes. Ultimately, we aim to uncover hidden value of enhanced reliable decision-making, whereby systems can make more informed, accurate and timely decisions, based on comprehensive data analytics and predictive insights. Mentioning selected ongoing cases, we highlight some benefits and challenges, like computational demand, data privacy concerns, and the need for transparency in AI decision-making. Underscoring the transformative implications of integrating FMs into the DT paradigm, a shift towards more intelligent, versatile and dynamic systems becomes clearer. We caution against the challenges of computational resources, safety considerations and interpretability. This step is pivotal towards unlocking unprecedented potential for advanced data-driven solutions in various industries.
Download
Paper Nr: 286
Title:

F4D: Factorized 4D Convolutional Neural Network for Efficient Video-Level Representation Learning
Authors:

Mohammad Al-Saad, Lakshmish Ramaswamy and Suchendra Bhandarkar
Abstract: Recent studies have shown that video-level representation learning is crucial to the capture and understanding of the long-range temporal structure for video action recognition. Most existing 3D convolutional neural network (CNN)-based methods for video-level representation learning are clip-based and focus only on short-term motion and appearances. These CNN-based methods lack the capacity to incorporate and model the long-range spatiotemporal representation of the underlying video and ignore the long-range video-level context during training. In this study, we propose a factorized 4D CNN architecture with attention (F4D) that is capable of learning more effective, finer-grained, long-term spatiotemporal video representations. We demonstrate that the proposed F4D architecture yields significant performance improvements over the conventional 2D, and 3D CNN architectures proposed in the literature. Experiment evaluation on five action recognition benchmark datasets, i.e., Something-Something-v1, Something-Something-v2, Kinetics-400, UCF101, and HMDB51 demonstrate the effectiveness of the proposed F4D network architecture for video-level action recognition.
Download
Paper Nr: 290
Title:

Explainable Deep Semantic Segmentation for Flood Inundation Mapping with Class Activation Mapping Techniques
Authors:

Jacob Sanderson, Hua Mao, Naruephorn Tengtrairat, Raid R. Al-Nima and Wai L. Woo
Abstract: Climate change is causing escalating extreme weather events, resulting in frequent, intense flooding. Flood inundation mapping is a key tool in com-bating these flood events, by providing insight into flood-prone areas, allowing for effective resource allocation and preparation. In this study, a novel deep learning architecture for the generation of flood inundation maps is presented and compared with several state-of-the-art models across both Sentinel-1 and Sentinel-2 imagery, where it demonstrates consistently superior performance, with an Intersection Over Union (IOU) of 0.5902 with Sentinel-1, and 0.6984 with Sentinel-2 images. The importance of this versatility is underscored by visual analysis of images from each satellite under different weather conditions, demonstrating the differing strengths and limitations of each. Explainable Artificial Intelligence (XAI) is leveraged to interpret the decision-making of the model, which reveals that the proposed model not only provides the greatest accuracy but exhibits an improved ability to confidently identify the most relevant areas of an image for flood detection.
Download
Paper Nr: 291
Title:

Is Amazon Kinesis Data Analytics Suitable as Core for an Event Processing Network Model?
Authors:

Arne Koschel, Irina Astrova, Anna Pakosch, Christian Gerner, Christin Schulze and Matthias Tyca
Abstract: This article looks at a proposed list of generalized requirements for a unified modelling of event processing networks (EPNs) and its application to Amazon Kinesis Data Analytics. It enhances our previous work in this area, in which we recently analyzed Apache Storm and earlier also the EPiA model, the BEMN model, and the RuleCore model. Our proposed EPN requirements look at both: The logical model of EPNs and the concrete technical implementation of them. Therefore, our article provides requirements for EPN models based on attributes derived from event processing in general as well as existing models. Moreover, as its core contribution, our article applies those requirements by an in depth analysis of Amazon Kinesis Data Analytics as a concrete implementation foundation of an EPN model.
Download
Paper Nr: 292
Title:

A Quantitative Assessment Framework for Modelling and Evaluation Using Representation Learning in Smart Agriculture Ontology
Authors:

Khadija Meghraoui, Teeradaj Racharak, Kenza A. El Kadi, Saloua Bensiali and Imane Sebari
Abstract: Understanding agricultural processes and their interactions can be improved with trustworthy and precise models. Such modelling boosts various related tasks, making it easier to take informed decisions in the realm of advanced agriculture. In our study, we present a novel agriculture ontology, primarily focusing on crop production. Our ontology captures fundamental domain knowledge concepts and their interconnections, particularly pertaining to key environmental factors. It encompasses static aspects like soil features, and dynamic ones such as climatic and thermal traits. In addition, we propose a quantitative framework for evaluating the quality of the ontology using the embeddings of all the concept names, role names, and individuals based on representation learning (i.e. OWL2Vec*, RDF2Vec, and Word2Vec) and dimensionality reduction for visualization (i.e. t-distributed Stochastic Neighbor Embedding). The findings validate the robustness of OWL2Vec* among other embedding algorithms in producing precise vector representations of ontology, and also demonstrate that our ontology has well-defined categorization aspects in conjunction of the embeddings.
Download
Paper Nr: 295
Title:

Sales Forecasting for Pricing Strategies Based on Time Series and Learning Techniques
Authors:

Jean-Christophe Ricklin, Ines Ben Amor, Raid Mansi, Vassilis Christophides and Hajer Baazaoui
Abstract: Time series exist in a wide variety of domains, such as market prices, healthcare and agriculture. Mod-elling time series data enables forecasting, anomaly detection, and data exploration. Few studies compare technologies and methodologies in the context of time series analysis, and existing tools are often limited in functionality. This paper focuses on the formulation and refinement of pricing strategies in mass retail, based on learning methods for sales forecasting and evaluation. The aim is to support BOOPER, a French startup specializing in pricing solutions for the retail sector. We focus on the strategy where each model is refined for a single product, studying both ensemble and parametric techniques as well as deep learning. To use these methods a hyperparameter setting is needed. The aim of this study is to provide an overview of the sensitivity of product sales to price fluctuations and promotions. The aim is also, to adapt existing methods using optimized machine and deep learning models, such as the Temporal Fusion Transformer (TFT) and the Temporal Convolutional Network (TCN), to capture the behaviour of each product. The idea is to improve their performance and adapt them to the specific requirements. We therefore provide an overview and experimental study of product learning models for each dataset, enabling informed decisions to be made about the most appropriate model and tool for each case.
Download
Paper Nr: 296
Title: