Publications

2022

• Pretext Tasks selection for multitask self-supervised speech representation learning
  
  • Zaiem Salah
  • Parcollet Titouan
  • Essid Slim
  • Heba Abdelwahab
  IEEE Journal of Selected Topics in Signal Processing, IEEE, 2022, 16 (6), pp.1439-1453. Through solving pretext tasks, self-supervised learning leverages unlabeled data to extract useful latent representations replacing traditional input features in the downstream task. In audio/speech signal processing, a wide range of features where engineered through decades of research efforts. As it turns out, learning to predict such features (a.k.a pseudo-labels) has proven to be a particularly relevant pretext task, leading to useful self-supervised representations which prove to be effective for downstream tasks. However, methods and common practices for combining such pretext tasks for better performance on the downstream task have not been explored and understood properly. In fact, the process relies almost exclusively on a computationally heavy experimental procedure, which becomes intractable with the increase of the number of pretext tasks. This paper introduces a method to select a group of pretext tasks among a set of candidates. The method we propose estimates calibrated weights for the partial losses corresponding to the considered pretext tasks during the self-supervised training process. The experiments conducted on automatic speech recognition, speaker and emotion recognition validate our approach, as the groups selected and weighted with our method perform better than classic baselines, thus facilitating the selection and combination of relevant pseudo-labels for self-supervised representation learning. (10.1109/JSTSP.2022.3195430)
  DOI : 10.1109/JSTSP.2022.3195430
• Solving X 2 3 n + 2 2 n + 2 n − 1 + ( X + 1 ) 2 3 n + 2 2 n + 2 n − 1 = b in F 2 4 n and an alternative proof of a conjecture on the differential spectrum of the related monomial functions
  
  • Kim Kwang Ho
  • Mesnager Sihem
  Finite Fields and Their Applications, Elsevier, 2022, 83, pp.102086. (10.1016/j.ffa.2022.102086)
  DOI : 10.1016/j.ffa.2022.102086
• Performance Analysis of CDL-impaired Multi-Core Fiber Transmission
  
  • Abouseif Akram
  • Damen Mohamed Oussama
  • Rekaya-Ben Othman Ghaya
  American Јournal of Optics and Photonics, 2022.
• The Jazz Ontology: A semantic model and large-scale RDF repositories for jazz
  
  • Proutskova Polina
  • Wolff Daniel
  • Fazekas György
  • Frieler Klaus
  • Höger Frank
  • Velichkina Olga
  • Solis Gabriel
  • Weyde Tillman
  • Pfleiderer Martin
  • Crayencour Hèlène Camille
  • Peeters Geoffroy
  • Dixon Simon
  Journal of Web Semantics, Elsevier, 2022, 74, pp.100735. (10.1016/j.websem.2022.100735)
  DOI : 10.1016/j.websem.2022.100735
• Is it Really Easy to Detect Sybil Attacks in C-ITS Environments: A Position Paper
  
  • Hammi Badis
  • Idir Yacine Mohamed
  • Zeadally Sherali
  • Khatoun Rida
  • Nebhen Jamel
  IEEE Transactions on Intelligent Transportation Systems, IEEE, 2022, 23 (10), pp.18273-18287. (10.1109/TITS.2022.3165513)
  DOI : 10.1109/TITS.2022.3165513
• SGD with Coordinate Sampling: Theory and Practice
  
  • Leluc Rémi
  • Portier François
  Journal of Machine Learning Research, Microtome Publishing, 2022, 23, pp.(342), pp.1-47. While classical forms of stochastic gradient descent algorithm treat the different coordinates in the same way, a framework allowing for adaptive (non uniform) coordinate sampling is developed to leverage structure in data. In a non-convex setting and including zeroth order gradient estimate, almost sure convergence as well as non-asymptotic bounds are established. Within the proposed framework, we develop an algorithm, MUSKETEER, based on a reinforcement strategy: after collecting information on the noisy gradients, it samples the most promising coordinate (all for one); then it moves along the one direction yielding an important decrease of the objective (one for all). Numerical experiments on both synthetic and real data examples confirm the effectiveness of MUSKETEER in large scale problems.
• 5G RAN : physical layer implementation and network slicing
  
  • de Javel Aymeric
  , 2022. A critical evolution from 4G to 5G is the heterogeneity of the terminals that connect the network. Those terminals range from smartphones to connected vehicles and sensors for agriculture. Given that the constraints and requirements associated with the different kinds of terminals are heterogeneous, it is not trivial to multiplex the services associated with them on top of a single physical infrastructure. Network slicing is the technology that enables the physical infrastructure to provide multiple logical networks (called network slices) to serve the various devices and associated services: this thesis studies network slicing and its implementation at the RAN level.One main issue raised by network slicing is resource allocation. Indeed, many models exist for resource allocation of the RAN but we are missing models which take into account new constraints implied by network slicing. The first contribution of this thesis is to define a new model for network slicing at the RAN level. This model takes into account diverse slices constraints such as capacity, UEs density, latency, and reliability. Simplicial homology is used to validate slices constraints fulfillment. Furthermore, this model is applied to power optimization, which is a critical aspect of network deployment. The second challenge addressed in this work is the network's supervision and control. Indeed, some verticals have ultra-high control requirements, and the network itself might not be able to satisfy this constraint fully. Therefore, we introduce a probe that can extract data from the network to feed supervision tools for the network's monitoring and control. This probe is designed to be resilient to cyber-attacks and is thus independent of the network.The last main contribution of this thesis is the introduction of an open-source 5G physical layer called free5GRAN. The physical layer provides all the minimal procedures and algorithms for communications between the gNodeB and UEs. The project's structure is built so one can easily modify it and implement new features. Furthermore, the software architecture is designed so that the physical layer is modular and can be derived to implement the open-RAN split 7.2.
• A machine learning based approach for the detection of sybil attacks in C-ITS
  
  • Hammi Badis
  • Idir Mohamed Yacine
  • Khatoun Rida
  , 2022, pp.1-4. (10.23919/APNOMS56106.2022.9919991)
  DOI : 10.23919/APNOMS56106.2022.9919991
• Fractionally Integrated Autoregressive Moving Average processes valued in a separable Hilbert space
  
  • Roueff François
  , 2022.
• Runtime Verification with Imperfect Information Through Indistinguishability Relations
  
  • Ferrando Angelo
  • Malvone Vadim
  , 2022, 13550, pp.335-351. Software systems are hard to trust, especially when autonomous. To overcome this, formal verification techniques can be deployed to verify such systems behave as expected. Runtime Verification is one of the most prominent and lightweight approaches to verify the system behaviour at execution time. However, standard Runtime Verification is built on the assumption of perfect information over the system, that is, the monitor checking the system can perceive everything. Unfortunately, this is not always the case, especially when the system under analysis contains rational/autonomous components and is deployed in real-world environments with possibly faulty sensors. In this work, we present an extension of the standard Runtime Verification of Linear Temporal Logic properties to consider scenarios with imperfect information. We present the engineering steps necessary to update the verification pipeline, and we report the corresponding implementation when applied to a case study involving robotic systems. (10.1007/978-3-031-17108-6_21)
  DOI : 10.1007/978-3-031-17108-6_21
• Alpha-capacity of communication channels with feedback: Theoretical overview
  
  • Rioul Olivier
  , 2022.
• Slice-aware Open Radio Access Network planning and dimensioning
  
  • Foroughi Parisa
  • Martins Philippe
  • Nivaggioli Patrice
  • Rougier Jean-Louis
  , 2022, pp.1-7. The fifth-generation (5G) of mobile networks and beyond is to host a variety of services for industry verticals with a diverse range of requirements. Network slicing (NS) is considered to be the fundamental enabling technology to address legacy networks' shortcoming, by tailoring logical virtual networks, called network slices, over the same infrastructure. Adopting the concepts of virtualization and open interfaces, Virtual radio access networks (vRAN) and Open RAN (ORAN) are two of the most promising architectures proposed for slicing radio access networks. To realize the efficient deployment (i.e. increase flexibility, scalability ,and decreased CAPEX and OPEX) of these architectures, a proper network planning approach is essential. This paper introduces a novel approach to planning and design of the ORAN architecture that takes into account QoS, CAPEX, OPEX and the transport network, simultaneously. The ORAN slice planning and design is formulated as a multi-objective optimization with binary variables and solved by simulated annealing.This paper provides a comprehensive discussion of the results. The proposed approach can be used in designing 5G ORAN network slices but also can be used as a transition network solution to integrate the 4G tier together with 5G for enabling a smooth and less costly transition. (10.1109/VTC2022-Fall57202.2022.10012946)
  DOI : 10.1109/VTC2022-Fall57202.2022.10012946
• Unexpectedness and Bayes’ Rule
  
  • Sileno Giovanni
  • Dessalles Jean-Louis
  , 2022, 13230, pp.107-116. (10.1007/978-3-031-12429-7_8)
  DOI : 10.1007/978-3-031-12429-7_8
• A Real-Time NMPC Controller for Autonomous Vehicle Racing
  
  • Li Nan
  • Goubault Eric
  • Pautet Laurent
  • Putot Sylvie
  , 2022, pp.148-155. Non-linear model predictive control (NMPC) solves structured optimization problems under predetermined constraints. It results in an optimal control series in a multiple-step prediction horizon. However, the NMPC requires considerable computation time, making it difficult to implement on devices with limited resources. We focus on an NMPC-based controller used for autonomous vehicle racing. It is a typical representative of quickly evolving cyber-physical systems. In the single-vehicle racing mode, we propose a triggering method to enable the execution of long-horizon NMPC, which is desirable for achieving a better lap time. For the head-to-head racing mode, in order to react rapidly to the evolving surroundings, we propose a short-horizon NMPC-based control strategy with safe overtaking capability. These control strategies can be implemented within a limited time budget. (10.1109/ICACR55854.2022.9935523)
  DOI : 10.1109/ICACR55854.2022.9935523
• Cache-Timing Attack on the SEAL Homomorphic Encryption Library
  
  • Cheng Wei
  • Danger Jean-Luc
  • Guilley Sylvain
  • Huang Fan
  • Bel Korchi Amina
  • Rioul Olivier
  , 2022. Homomorphic encryption (HE) ensures provable secrecy of data processed in the ciphertext domain. However, it happens that FHE private-key algorithms can be broken by side-channel attacks. We disclose a novel cache-timing attack on the SEAL open-source HE library. It is triggered by a non-constant time Barrett modular multiplication, which is one of the building blocks in SEAL. We both analyze the mathematical conditions upon which the leakage occurs and show the experimental feasibility of the attack.
• Identity-Based Encryption from the Tate Pairing on Genus Two Curves
  
  • Zitouni Mohammed
  • Guilley Sylvain
  • Mokrane Farid
  , 2022. Identity Based Encryption is an approach to link the public key to an identity. It is an extremely useful asymmetric cryptography type in which public and private keys are computed from a known identifier such as an email address instead of being generated randomly. This allows more flexibility in managing ad-hoc public key encryption and ensuring secure communications. The aim of this work is to improve IBE scheme using the bilinear Tate pairing on genus two curves with ordinary Jacobian over large prime fields. We present a full description of functional IBE scheme using the optimization of the Tate pairing computations. The proposed application answers a question of Boneh and Franklin [2] about the possibility of using the Tate pairing in IBE schemes and represents the first IBE exploiting pairings in genus two. We provide a full description of a functional IBE scheme using the optimization of the Tate pairing computations.
• Neural Belief Propagation Auto-Encoder for Linear Block Code Design
  
  • Larue Guillaume
  • Dufrene Louis-Adrien
  • Lampin Quentin
  • Ghauch Hadi
  • Othman Ghaya Rekaya-Ben
  IEEE Transactions on Communications, Institute of Electrical and Electronics Engineers, 2022, 70 (11), pp.7250-7264. The growing number of Internet of Thing (IoT) and Ultra-Reliable Low Latency Communications (URLCC) use cases in next generation communication networks calls for the development of efficient Forward Error Correction (FEC) mechanisms. These use cases usually imply using short to mid-sized information blocks and requires low-complexity and/or fast decoding procedures. This paper investigates the joint learning of short to mid block-length coding schemes and associated Belief-Propagation (BP) like decoders using Machine Learning (ML) techniques. An interpretable auto-encoder (AE) architecture is proposed, ensuring scalability to block sizes currently challenging for ML-based linear block code design approaches. By optimizing a coding scheme w.r.t. the targeted decoder, the proposed system offers a good complexity/performance trade-off compared to various codes from literature with length up to 128 bits. (10.1109/TCOMM.2022.3208331)
  DOI : 10.1109/TCOMM.2022.3208331
• The Impact of Multi-scale Control Topology on Asset Distribution in Dynamic Environments
  
  • Zahadat Payam
  • Diaconescu Ada
  , 2022, pp.31-36. In many self-organising systems the ability to extract necessary resources from the external environment is essential for growth and survival. E.g., extracting sunlight and nutrients in organic plants, monetary income in business organisations and mobile robots in intelligent swarms. When operating within competitive, changing environments, such systems must distribute their assets wisely, to improve and adapt their ability to extract available resources. As the system size increases, the assetdistribution process often gets organised around a multi-scale control topology. This topology may be static (fixed) or dynamic (enabling growth and structural adaptation) depending on the system’s constraints and adaptive mechanisms. In this paper we expand on a plant-inspired asset-distribution model and study the impact that the topology of the multi-scale control process has upon the system’s ability to self-adapt asset distribution when resource availability changes within the environment. Results show how different topological characteristics and different competition levels between system branches impact overall system profitability, adaptation delays and disturbances when environmental changes occur. These findings provide a basis for system designers to select the most suitable topology and configuration for their particular application and execution environment. (10.1109/ACSOSC56246.2022.00023)
  DOI : 10.1109/ACSOSC56246.2022.00023
• Software Artifact Mining in Software Engineering Conferences: A Meta-Analysis
  
  • Abou Khalil Zeinab
  • Zacchiroli Stefano
  , 2022. Background: Software development results in the production of various types of artifacts: source code, version control system metadata, bug reports, mailing list conversations, test data, etc. Empirical software engineering (ESE) has thrived mining those artifacts to uncover the inner workings of software development and improve its practices. But which artifacts are studied in the field is a moving target, which we study empirically in this paper. Aims: We quantitatively characterize the most frequently mined and co-mined software artifacts in ESE research and the research purposes they support. Method: We conduct a meta-analysis of artifact mining studies published in 11 top conferences in ESE, for a total of 9621 papers. We use natural language processing (NLP) techniques to characterize the types of software artifacts that are most often mined and their evolution over a 16-year period (2004-2020). We analyze the combinations of artifact types that are most often mined together, as well as the relationship between study purposes and mined artifacts. Results: We find that: (1) mining happens in the vast majority of analyzed papers, (2) source code and test data are the most mined artifacts, (3) there is an increasing interest in mining novel artifacts, together with source code, (4) researchers are most interested in the evaluation of software systems and use all possible empirical signals to support that goal. (10.1145/3544902.3546239)
  DOI : 10.1145/3544902.3546239
• Mitigating Gender Bias of Pre-Trained Face Recognition Models with an Ethical Module
  
  • Conti Jean-Rémy
  • Noiry Nathan
  • Despiegel Vincent
  • Gentric Stéphane
  • Clémençon Stephan
  , 2022. In spite of the high performance and reliability of deep learning algorithms in a wide range of everyday applications, many investigations tend to show that a lot of models exhibit biases, discriminating against specific subgroups of the population (e.g. gender, ethnicity). This urges the practitioner to develop fair systems with a uniform/comparable performance across sensitive groups. In this work, we investigate the gender bias of deep Face Recognition networks. In order to measure this bias, we introduce two new metrics, BFAR and BFRR, that better reflect the inherent deployment needs of Face Recognition systems. Motivated by geometric considerations, we mitigate gender bias through a new post-processing methodology which transforms the deep embeddings of a pre-trained model to give more representation power to discriminated subgroups. It consists in training a shallow neural network by minimizing a Fair von Mises-Fisher loss whose hyperparameters account for the intra-class variance of each gender. Interestingly, we empirically observe that these hyperparameters are correlated with our fairness metrics. In fact, a careful selection significantly reduces gender bias. This paper, in its previous form, has been accepted at ICML 2022.
• Lensing: It is All About Perspective
  
  • Diaconescu Ada
  • King David
  • Bellman Kirstie
  • Landauer Christopher
  • Nelson Phyllis
  , 2022, ACSOS-C 2022, pp.119-125. Complex Adaptive Systems (CAS) are difficult to understand and predict. Certain CAS phenomena can only be observed at certain abstraction levels – e.g., swarm dynamics cannot be analysed by only tracking one individual. In most current systems the “right” abstraction levels for each question asked, or each problem solved, are determined at design time. Yet, in adaptive self-integrating systems the abstraction level must be determined dynamically, depending on the questions encountered and the system context. This position paper aims to set-up the basis for a research initiative in this direction. It proposes the concept of ‘lensing’ to tune a system’s observation granularity in terms of spatial and temporal scope, and information detail. It further introduces lens efficacy and efficiency to evaluate a lens’ ability to answer specific questions–a critical process for self-aware systems performing lensing at runtime. Finally, the paper illustrates the criticality of lensing in answering different kinds of questions via several examples of collective movement systems, including a Game of Life (GoL) Glider simulation. (10.1109/ACSOSC56246.2022.00044)
  DOI : 10.1109/ACSOSC56246.2022.00044
• A generic and modular reference architecture for self-explainable smart homes
  
  • Houze Etienne
  • Diaconescu Ada
  • Dessalles Jean-Louis
  • Menga David
  , 2022, ACSOS 2022, pp.101-110. Explainable AI (XAI) has become a major topic in Artificial Intelligence since the mid 2010s. While smart home explainability promises to improve user experience and trust, it is mostly left outside the scope of current AI research. We identify three main challenges that may cause this delay. First, smart device heterogeneity hinders the development of a system-wide vocabulary and communication medium required for end-to-end explanation. Second, smart home runtime changes – e.g. dynamic component additions, deletions and updates – require corresponding explanatory updates. Third, explanation context runtime changes: word meanings may vary with the end-user and the passing seasons – e.g. the notion of cold may vary, depending on the month. To tackle these challenges, we propose a generic, modular XAI reference architecture featuring: i) Local Explanatory Components (LECs) that provide resource-specific explanatory expertise and support runtime extensions; ii) mapping capabilities that allow LECs to translate resource-specific monitoring variables into resource-independent abstractions – predicates and events – which can then be used for generic inter-LEC communication; iii) a generic central component, called Spotlight, that coordinates LECs to generate system-wide explanations. We validate our proposal via a cyber-physical prototype of self-explainable smart home, implemented via a physical home maquette equipped with GrovePi sensors. We show how our prototype can handle several realistic scenarios highlighting the main issues identified above. This provides an initial stepping-stone towards a fully self-explanatory smart home solution. The genericity of our proposal opens the way for transferring it to similar application domains. (10.1109/ACSOS55765.2022.00028)
  DOI : 10.1109/ACSOS55765.2022.00028
• Monte Carlo Tree Search Bidding Strategy for Simultaneous Ascending Auctions
  
  • Pacaud Alexandre
  • Coupechoux Marceau
  • Bechler Aurelien
  , 2022, pp.322-329. We tackle in this work the problem for a player to efficiently bid in Simultaneous Ascending Auctions (SAA). Although the success of SAA partially comes from its relative simplicity, bidding efficiently in such an auction is complicated as it presents a number of complex strategical problems. No generic algorithm or analytical solution has yet been able to compute the optimal bidding strategy in face of such complexities. By modelling the auction as a turn-based deterministic game with complete information, we propose the first algorithm which tackles simultaneously two of its main issues: exposure and own price effect. Our bidding strategy is computed by Monte Carlo Tree Search (MCTS) which relies on a new method for the prediction of closing prices. We show that our algorithm significantly outperforms state-of-the-art existing bidding methods. More precisely, our algorithm achieves a higher expected utility by taking lower risks than existing strategies. (10.23919/WiOpt56218.2022.9930539)
  DOI : 10.23919/WiOpt56218.2022.9930539
• ORSUM 2022 - 5th Workshop on Online Recommender Systems and User Modeling
  
  • Vinagre João
  • Al-Ghossein Marie
  • Jorge Alípio Mário
  • Bifet Albert
  • Peska Ladislav
  , 2022, pp.661--662. Modern online systems for user modeling and recommendation need to continuously deal with complex data streams generated by users at very fast rates. This can be overwhelming for systems and algorithms designed to train recommendation models in batches, given the continuous and potentially fast change of content, context and user preferences or intents. Therefore, it is important to investigate methods able to transparently and continuously adapt to the inherent dynamics of user interactions, preferably for long periods of time. Online models that continuously learn from such flows of data are gaining attention in the recommender systems community, given their natural ability to deal with data generated in dynamic, complex environments. User modeling and personalization can particularly benefit from algorithms capable of maintaining models incrementally and online. The objective of this workshop is to foster contributions and bring together a growing community of researchers and practitioners interested in online, adaptive approaches to user modeling, recommendation and personalization, and their implications regarding multiple dimensions, such as evaluation, reproducibility, privacy, fairness and transparency. (10.1145/3523227.3547411)
  DOI : 10.1145/3523227.3547411
• Modeling Rowhammer in the gem5 simulator
  
  • France Loïc
  • Bruguier Florent
  • Mushtaq Maria
  • Novo David
  • Benoit Pascal
  , 2022.