Publications

2023

• Towards efficient, general and robust entity disambiguation systems
  
  • Chen Lihu
  , 2023. Entity disambiguation aims to map mentions in documents to standard entities in a given knowledge base, which is important for various applications such as information extraction, Web search and question answering.Although the field is very vibrant with many novel works popping up, there are three questions that are underexplored by prior work.1) Can we use a small model to approach the performance of a big model?2) How to develop a single disambiguation system adapted to multiple domains?3) Are existing systems robust to out-of-vocabulary words and different word orderings?Based on the three questions, we explore how to construct an efficient, general and robust entity disambiguation system. We also successfully apply entity disambiguation to the knowledge base completion task, especially for the long-tail entities.
• Preventing Discriminatory Decision-making in Evolving Data Streams
  
  • Wang Zichong
  • Saxena Nripsuta
  • Yu Tongjia
  • Karki Sneha
  • Zetty Tyler
  • Haque Israat
  • Zhou Shan
  • Kc Dukka
  • Stockwell Ian
  • Wang Xuyu
  • Bifet Albert
  • Zhang Wenbin
  , 2023, pp.149--159. Bias in machine learning has rightly received significant attention over the past decade. However, most fair machine learning (fair-ML) works to address bias in decision-making systems has focused solely on the offline setting. Despite the wide prevalence of online systems in the real world, work on identifying and correcting bias in the online setting is severely lacking. The unique challenges of the online environment make addressing bias more difficult than in the offline setting. First, Streaming Machine Learning (SML) algorithms must deal with the constantly evolving real-time data stream. Secondly, they need to adapt to changing data distributions (concept drift) to make accurate predictions on new incoming data. Incorporating fairness constraints into this already intricate task is not straightforward. In this work, we focus on the challenges of achieving fairness in biased data streams while accounting for the presence of concept drift, accessing one sample at a time. We present Fair Sampling over Stream (FS2), a novel fair rebalancing approach capable of being integrated with SML classification algorithms. Furthermore, we devise the first unified performance-fairness metric, Fairness Bonded Utility (FBU), to efficiently evaluate and compare the trade-offs between performance and fairness across various bias mitigation methods. FBU simplifies the comparison of fairness-performance trade-offs of multiple techniques through one unified and intuitive evaluation, allowing model designers to easily choose a technique. Overall, extensive evaluations show our measures surpass those of other fair online techniques previously reported in the literature. (10.1145/3593013.3593984)
  DOI : 10.1145/3593013.3593984
• Toward Robust Analog Equilibrium Propagation by Investigating Power Calculation Dynamics
  
  • Kiraz Fatma Zulal
  • Pham Dang-Kièn Germain
  • Desgreys Patricia
  , 2023. In this work, we investigate the robustness and consistency of the Equilibrium Propagation (EqProp) algorithm for training analog neural networks, addressing previous research limitations. We analyze the algorithm's robustness concerning learning parameter variations and examine the relationship between power calculations and convergence behavior. By accounting for all relevant components and their impact on learning, we aim to develop a reliable and consistent algorithm based on EqProp principles.
• Neural Network based 5G Power Amplifier Modeling on MATLAB
  
  • Pham Thuy T.
  • Pham Dang-Kièn G.
  • Desgreys Patricia
  • Bouazza Tayeb H. C.
  , 2023.
• Introduction to the Special Issue: 5+G Network Energy Consumption, Energy Efficiency and Environmental Impact
  
  • Ware Cédric
  • Coupechoux Marceau
  • Hossain Ekram
  • Mas-Machuca Carmen
  • Sharma Vinod
  • Tzanakaki Anna
  Annals of Telecommunications - annales des télécommunications, Springer, 2023. (10.1007/s12243-023-00967-6)
  DOI : 10.1007/s12243-023-00967-6
• Questioning the ability of feature-based explanations to empower non-experts in robo-advised financial decision-making
  
  • Bertrand Astrid
  • Eagan James
  • Maxwell Winston
  , 2023, pp.943-958. Robo-advisors are democratizing access to life-insurance by enabling fully online underwriting. In Europe, financial legislation requires that the reasons for recommending a life insurance plan be explained according to the characteristics of the client, in order to empower the client to make a "fully informed decision". In this study conducted in France, we seek to understand whether legal requirements for feature-based explanations actually help users in their decision-making. We conduct a qualitative study to characterize the explainability needs formulated by non-expert users and by regulators expert in customer protection. We then run a large-scale quantitative study using Robex, a simplified robo-advisor built using ecological interface design that delivers recommendations with explanations in different hybrid textual and visual formats: either "dialogic"-more textual-or "graphical"-more visual. We find that providing feature-based explanations does not improve appropriate reliance or understanding compared to not providing any explanation. In addition, dialogic explanations increase users' trust in the recommendations of the robo-advisor, sometimes to the users' detriment. This real-world scenario illustrates how XAI can address information asymmetry in complex areas such as finance. This work has implications for other critical, AI-based recommender systems, where the General Data Protection Regulation (GDPR) may require similar provisions for feature-based explanations. CCS CONCEPTS • Human-centered computing → Empirical studies in HCI. (10.1145/3593013.3594053)
  DOI : 10.1145/3593013.3594053
• A generic information-theoretic framework for evaluating the side-channel security of masked implementations
  
  • Rioul Olivier
  • Béguinot Julien
  • Liu Yi
  • Cheng Wei
  • Guilley Sylvain
  , 2023. We propose an information-theoretic framework that aims at unifying and optimizing several previous works on the side-channel security of masked im- plementations of any order d in some Abelian group: Duc at al.at EURO- CRYPT2015, Dziembowski et al.at TCC2016, Ch ́erisey et al.at CHES2019, Prest et al.at CRYPTO2019, Masure et al.at CARDIS2022, Ito et al.at CCS2022, Liu et al.at ITW2023, and B ́eguinot et al.at COSADE2023 and ISIT2023. In this general framework, two theoretical ingredients are system- atically leveraged: (i) a variation of a Fano inequality relating the attack performance (success rate) to a measure of information between the sensi- tive variable and the leakage; (ii) a variation of a Mrs. Gerber Lemma lower bounding a statistical measure of the sensitive variable by the product of sim- ilar measures for its d + 1 masking shares. Depending on the choice of the information measure and of the statistical measure, and possibly on Pinsker- type inequalities relating these measures, one can establish anew all previously published lower bounds on the number of queries necessary to achieve a given attack success rate. These results make progress on the evaluation of the secu- rity guarantees of higher order masking, and stimulate further research on best possible bounds & possible application to other types of masking schemes.
• Science and Engineering for What? A Large-Scale Analysis of Students' Projects in Science Fairs
  
  • Eloy Adelmo
  • Palmeira Ferraz Thomas
  • Silva Alves Fellip
  • Deus Lopes Roseli De
  , 2023, pp.946-949. Science and Engineering fairs offer K-12 students opportunities to engage with authentic STEM practices. Particularly, students are given the chance to experience authentic and open inquiry processes, by defining which themes, questions and approaches will guide their scientific endeavors. In this study, we analyzed data from over 5,000 projects presented at a nationwide science fair in Brazil over the past 20 years using topic modeling to identify the main topics that have driven students' inquiry and design. Our analysis identified a broad range of topics being explored, with significant variations over time, region, and school setting. We argue those results and proposed methodology can not only support further research in the context of science fairs, but also inform instruction and design of contexts-specific resources to support students in open inquiry experiences in different settings.
• Attack Graphs & Subset Sabotage Games
  
  • Catta Davide
  • Leneutre Jean
  • Malvone Vadim
  Intelligenza Artificiale, IOS press, 2023, 17 (1), pp.77-88. We consider an extended version of sabotage games played over Attack Graphs. Such games are two-player zero-sum reachability games between an Attacker and a Defender. This latter player can erase particular subsets of edges of the Attack Graph. To reason about such games we introduce a variant of Sabotage Modal Logic (that we call Subset Sabotage Modal Logic) in which one modality quantifies over non-empty subset of edges. We show that we can characterize the existence of winning Attacker strategies by formulas of Subset Sabotage Modal Logic. (10.3233/IA-221080)
  DOI : 10.3233/IA-221080
• Sur la faisabilité d’une compensation efficace de la latence en utilisant l’extrapolation des images vidéo
  
  • Kanj Hind
  • Trioux Anthony
  • Cagnazzo Marco
  • Coudoux François-Xavier
  • Corlay Patrick
  • Kieffer M.
  , 2023.
• Pairing Method
  
  • Khalfaoui Sameh
  • Villard Arthur
  • MA Jingxuan
  • Leneutre Jean
  , 2023.
• SeqDQN: Multi-Agent Deep Reinforcement Learning for Uplink URLLC with Strict Deadlines
  
  • Robaglia Benoît-Marie
  • Coupechoux Marceau
  • Tsilimantos Dimitrios
  • Destounis Apostolos
  , 2023. Recent studies suggest that Multi-Agent Reinforcement Learning (MARL) can be a promising approach to tackle wireless telecommunication problems and Multiple Access (MA) in particular. The most relevant MARL algorithms for distributed MA are those with "decentralized execution", where an agent's actions are only functions of their own local observation history and agents cannot exchange any information. Centralized-Training-Decentralized-Execution (CTDE) and Independent Learning (IL) are the two main families in this category. However, while the former suffers from high communication overhead during the centralized training, the latter suffers from various theoretical shortcomings. In this paper, we first study the performance of these two MARL frameworks in the context of Ultra Reliable Low Latency Communication (URLLC), where MA is constrained by strict deadlines. Second, we propose a new distributed MARL framework, namely SeqDQN, leveraging the constraints of our URLLC problem to train agents in a more efficient way. We demonstrate that not only does our solution outperform the traditional random access baselines, but it also outperforms state-of-the-art MARL algorithms in terms of performance and convergence time.
• Participation de l'équipe TTGV à DEFT 2023~: Réponse automatique à des QCM issus d'examens en pharmacie
  
  • Blivet Andréa
  • Degrutère Solène
  • Gendron Barbara
  • Renault Aurélien
  • Siouffi Cyrille
  • Gaudray Bouju Vanessa
  • Cerisara Christophe
  • Flamein Hélène
  • Guibon Gaël
  • Labeau Matthieu
  • Rousseau Tom
  , 2023, pp.23-38. Cet article présente l'approche de l'équipe TTGV dans le cadre de sa participation aux deux tâches proposées lors du DEFT 2023 : l'identification du nombre de réponses supposément justes à un QCM et la prédiction de l'ensemble de réponses correctes parmi les cinq proposées pour une question donnée. Cet article présente les différentes méthodologies mises en oeuvre, explorant ainsi un large éventail d'approches et de techniques pour aborder dans un premier temps la distinction entre les questions appelant une seule ou plusieurs réponses avant de s'interroger sur l'identification des réponses correctes. Nous détaillerons les différentes méthodes utilisées, en mettant en exergue leurs avantages et leurs limites respectives. Ensuite, nous présenterons les résultats obtenus pour chaque approche. Enfin, nous discuterons des limitations intrinsèques aux tâches elles-mêmes ainsi qu'aux approches envisagées dans cette contribution.
• Protocole d'annotation multi-label pour une nouvelle approche à la génération de réponse socio-émotionnelle orientée-tâche
  
  • Vanel Lorraine
  • Yacoubi Alya
  • Clavel Chloe
  , 2023, pp.335-348. Depuis l'apparition des systèmes conversationnels, la modélisation des comportements humains constitue un axe de recherche majeur afin de renforcer l'expression des attributs émotionnels de ces systèmes. En nous intéressant aux agents conversationnels génératifs orientés-tâches, nous proposons une nouvelle approche pour rendre la réponse générée plus pertinente au contexte émotionnel de l'interlocuteur. Cette approche consiste à ajouter une étape supplémentaire de prédiction de labels pour conditionner la réponse générée et assurer sa pertinence au contexte socio-émotionnel de l'utilisateur. Nous proposons une formulation de cette nouvelle tâche de prédiction en nous appuyant sur un protocole d'annotation de données que nous avons conçu et implémenté. À travers cet article, nous apportons les contributions suivantes: la formulation de la tâche de prédiction de labels socio-émotionnels et la description du protocole d'annotation associé. Avec cette méthodologie, nous visons à développer des systèmes conversationnels socialement pertinents et indépendants.
• Un mot, deux facettes : traces des opinions dans les représentations contextualisées des mots
  
  • Garí Soler Aina
  • Labeau Matthieu
  • Clavel Chloe
  , 2023, pp.49-57. La façon dont nous utilisons les mots est influencée par notre opinion. Nous cherchons à savoir si cela se reflète dans les plongements de mots contextualisés. Par exemple, la représentation d' « animal » est-elle différente pour les gens qui voudraient abolir les zoos et ceux qui ne le voudraient pas ? Nous explorons cette question du point de vue du changement sémantique des mots. Nos expériences avec des représentations dérivées d'ensembles de données annotés avec les points de vue révèlent des différences minimes, mais significatives, entre postures opposées.
• Aging-Induced Failure Prognosis via Digital Sensors
  
  • Anik Md Toufiq Hasan
  • Reefat Hasin Ishraq
  • Danger Jean-Luc
  • Guilley Sylvain
  • Karimi Naghmeh
  , 2023, pp.703-708. Aggressive scaling continues to push technology into smaller feature sizes and results in more complex systems in a single chip. With such scaling, various robustness concerns have come into account among which the change of circuits' properties during their lifetime, so-called device aging, has received a lot of attention. Due to aging, the electrical behavior of transistors deviates from its original intended one resulting in degrading the chip's performance, and ultimately the chip fails to provide correct outputs. Thereby, prognosis of circuit performance degradation during the runtime, before the chip actually fails is highly crucial in increasing the reliability of chips. Accordingly in this paper, we develop a machine-learning based framework that, leveraging the outcome of embedded time-to-digital-convertors (so-called "digital sensors''), predicts aging-induced degradation. This information can be used to prevent chip failures via deploying Dynamic Voltage and Frequency Scaling (DVFS) (10.1145/3583781.3590204)
  DOI : 10.1145/3583781.3590204
• The Long Road to Sobriety: Estimating the Operational Power Consumption of Cellular Base Stations in France
  
  • Ahmed Arsalan
  • Coupechoux Marceau
  , 2023. As the Information and Communication Technology (ICT) sector represents 1.8% to 3.9% of the global Green House Gas (GHG) emissions, it is of upmost importance to know how much energy is spent annually in mobile networks and how this consumption is evolving. It is quite likely that the huge energy efficiency gains achieved by technology evolution have at least been compensated by the surge in data traffic. Therefore, in this paper, we estimate the operational power consumption of cellular Base Stations (BSs) deployed in France from 2015 to 2022. However, unfortunately, the lack of openly available data hinders the estimation process. In order to work around this issue, we rely on a public dataset on radio electric installations, on widely adopted power consumption models and on a set of assumptions backed by the scientific literature. We demonstrate that, over the considered period, the numbers of BSs and transceivers have grown at a sustained Compound Annual Growth Rate (CAGR) of 7.55% and 18.27%, respectively. Within the same period, the average BS power consumption has increased at a CAGR of 9.89% while the total operational power consumption of BSs has grown at a CAGR of 18.18%. We further show that the introduction of 5G has accelerated this trend despite the recent decommissioning of 2G and 3G transceivers. These alarming figures advocate for proactive digital sobriety policies.
• Cosmopolite Sound Monitoring (CoSMo): A Study of Urban Sound Event Detection Systems Generalizing to Multiple Cities
  
  • Angulo Florian
  • Essid Slim
  • Peeters Geoffroy
  • Mietlicki Christophe
  , 2023, pp.1-5. Measuring noise in cities and automatically identifying the corresponding sound sources are a crucial challenge for policymakers. Indeed, such information helps addressing noise pollution and improving the well-being of urban dwellers. In recent years, researchers have provided annotated datasets recorded in two major cities to foster the development of urban sound event detection (SED) systems. This paper presents an in-depth study of the behaviour of state-of-the-art SED systems well suited to our problem, combining three far-field real recordings datasets which can be used jointly during training. In our evaluation, we highlight the performance gaps existing between simple and hard recording examples based on the salience of sound events and the polyphony of the recordings. We provide new proximity annotations for this analysis. We evaluate the ability of urban SED systems to generalize across cities with varying degrees of training supervision. We show that such generalization is hindered mostly by the difficulties current urban SED systems have to detect sound events with low salience along with sound events in highly polyphonic soundscapes. (10.1109/ICASSP49357.2023.10095833)
  DOI : 10.1109/ICASSP49357.2023.10095833
• Time-varying Signals Recovery via Graph Neural Networks
  
  • Castro-Correa John A
  • Giraldo Jhony H
  • Mondal Anindya
  • Badiey Mohsen
  • Bouwmans Thierry
  • Malliaros Fragkiskos D.
  , 2023, pp.1-5. The recovery of time-varying graph signals is a fundamental problem with numerous applications in sensor networks and forecasting in time series. Effectively capturing the spatio-temporal information in these signals is essential for the downstream tasks. Previous studies have used the smoothness of the temporal differences of such graph signals as an initial assumption. Nevertheless, this smoothness assumption could result in a degradation of performance in the corresponding application when the prior does not hold. In this work, we relax the requirement of this hypothesis by including a learning module. We propose a Time Graph Neural Network (TimeGNN) for the recovery of time-varying graph signals. Our algorithm uses an encoder-decoder architecture with a specialized loss composed of a mean squared error function and a Sobolev smoothness operator.TimeGNN shows competitive performance against previous methods in real datasets. (10.1109/ICASSP49357.2023.10096168)
  DOI : 10.1109/ICASSP49357.2023.10096168
• Higher-order Sparse Convolutions in Graph Neural Networks
  
  • Giraldo Jhony H.
  • Javed Sajid
  • Mahmood Arif
  • Malliaros Fragkiskos D.
  • Bouwmans Thierry
  , 2023, pp.1-5. Graph Neural Networks (GNNs) have been applied to many problems in computer sciences. Capturing higher-order relationships between nodes is crucial to increase the expressive power of GNNs. However, existing methods to capture these relationships could be infeasible for large-scale graphs. In this work, we introduce a new higher-order sparse convolution based on the Sobolev norm of graph signals. Our Sparse Sobolev GNN (S-SobGNN) computes a cascade of filters on each layer with increasing Hadamard powers to get a more diverse set of functions, and then a linear combination layer weights the embeddings of each filter. We evaluate S-SobGNN in several applications of semi-supervised learning. S-SobGNN shows competitive performance in all applications as compared to several state-of-the-art methods. (10.1109/ICASSP49357.2023.10096494)
  DOI : 10.1109/ICASSP49357.2023.10096494
• Explainable Audio Classification of Playing Techniques with Layer-wise Relevance Propagation
  
  • Wang Changhong
  • Lostanlen Vincent
  • Lagrange Mathieu
  , 2023, pp.1-5. Deep convolutional networks (convnets) in the time-frequency domain can learn an accurate and fine-grained categorization of sounds. For example, in the context of music signal analysis, this categorization may correspond to a taxonomy of playing techniques: vibrato, tremolo, trill, and so forth. However, convnets lack an explicit connection with the neurophysiological underpinnings of musical timbre perception. In this article, we propose a data-driven approach to explain audio classification in terms of physical attributes in sound production. We borrow from current literature in "explainable AI" (XAI) to study the predictions of a convnet which achieves an almost perfect score on a challenging task: i.e., the classification of five comparable real-world playing techniques from 30 instruments spanning seven octaves. Mapping the signal into the carrier-modulation domain using scattering transform, we decompose the networks' predictions over this domain with layer-wise relevance propagation. We find that regions highly-relevant to the predictions localized around the physical attributes with which the playing techniques are performed. (10.1109/ICASSP49357.2023.10095894)
  DOI : 10.1109/ICASSP49357.2023.10095894
• LEARNING INTERPRETABLE FILTERS IN WAV-UNET FOR SPEECH ENHANCEMENT
  
  • Mathieu Félix
  • Courtat Thomas
  • Richard Gael
  • Peeters Geoffroy
  , 2023. Due to their performances, deep neural networks have emerged as a major method in nearly all modern audio processing applications. Deep neural networks can be used to estimate some parameters or hyperparameters of a model, or in some cases the entire model in an end-to-end fashion. Although deep learning can lead to state of the art performances, they also suffer from inherent weaknesses as they usually remain complex and non interpretable to a large extent. For instance, the internal filters used in each layers are chosen in an adhoc manner with only a loose relation with the nature of the processed signal. We propose in this paper an approach to learn interpretable filters within a specific neural architecture which allow to better understand the behaviour of the neural network and to reduce its complexity. We validate the approach on a task of speech enhancement and show that the gain in interpretability does not degrade the performance of the model.
• Fine-tuning strategies for faster inference using speech self-supervised models: a comparative study
  
  • Zaiem Salah
  • Algayres Robin
  • Parcollet Titouan
  • Essid Slim
  • Ravanelli Mirco
  , 2023. Self-supervised learning (SSL) has allowed substantial progress in Automatic Speech Recognition (ASR) performance in low-resource settings. In this context, it has been demonstrated that larger selfsupervised feature extractors are crucial for achieving lower downstream ASR error rates. Thus, better performance might be sanctioned with longer inferences. This article explores different approaches that may be deployed during the fine-tuning to reduce the computations needed in the SSL encoder, leading to faster inferences. We adapt a number of existing techniques to common ASR settings and benchmark them, displaying performance drops and gains in inference times. Interestingly, we found that given enough downstream data, a simple downsampling of the input sequences outperforms the other methods with both low performance drops and high computational savings, reducing computations by 61.3% with an WER increase of only 0.81. Finally, we analyze the robustness of the comparison to changes in dataset conditions, revealing sensitivity to dataset size.
• Multi layered Misbehavior Detection for a connected and autonomous vehicle
  
  • Bouchouia Mohammed
  , 2023. In recent years, the vehicular field has undergone significant advancements with the development of autonomous vehicles and smart cities. These advancements have brought about a modernization of human life, where everything is interconnected - from individuals through smartphones to infrastructure, cars, and motorcycles. In such a system, information is exchanged and processed, and used to ensure the proper functioning of all entities. However, the increased reliance on V2X communication also makes it a target for security attacks, which could lead to the dissemination of false or manipulated information from malicious sources. This could pose a threat to the proper functioning of the system and can potentially result in accidents. To address this problem, it is crucial to validate and verify the communication to ensure its accuracy and prevent malicious attacks. We aim to formulate misbehavior and misbehavior detection for connected and autonomous vehicles of level 4/5 automation. In our thesis, we propose a multi-layered architecture for the detection of abnormal behaviors with automatic learning to secure the connected and autonomous vehicles' communications, sensors, and internal components. The architecture allows us to propose a novel reinforcement learning based neural architecture for the detection of misbehaviors where we showed in a simulated environment, through evaluation, that the model is capable of detecting novel misbehaviors and performs better than current state-of-the-art algorithms. Furthermore, we tackle data leakage in V2X data and propose a cross-validation method to avoid said leakage in machine learning applications. We also developed a simulation for vehicular environments capable of injecting and detecting misbehaviors for the evaluation of our thesis results. The ideas developed in this research have resulted in several publications and have the potential to significantly enhance the security and reliability of vehicular systems.
• Procédé et système d’authentification par un équipement vérificateur d’un dispositif à authentifier équipé d’un circuit PUF
  
  • Khalfaoui Sameh
  • Villard Arthur
  • Ma Jingxuan
  • Leneutre Jean
  , 2023.