Publications

2023

• Fault-Tolerant Four-Dimensional Constellation for Coherent Optical Transmission Systems
  
  • Liu Jingtian
  • Awwad Élie
  • Jaouën Yves
  , 2023. We propose a 4-dimensional 2-ary amplitude ring-switched modulation format with 64 symbols, which is denoted as 4D-2A-RS64 encoded over two polarization tributaries to improve the transmission performance over long-haul optical fibers in the presence of the non-linear Kerr effect. At a spectral efficiency of 6 bits per 4D, simulation results show that this format outperforms the polarization division multiplexed (PDM) 8QAM-star modulation as well as the 4D-2A-8PSK over links without inline dispersion management. We evaluate the performance for a WDM transmission of 11 × 90 Gbaud channels over a multi-span SSMF link. For an achievable information rate of 4.8bit/s/Hz, the maximum transmission distance is improved by 10.6% (400 km) and 4% (160 km) compared to PDM-8QAM-star and 4D-2A-8PSK respectively. The achieved gains are composed of a linear part and a non-linear part, respectively from the improved Euclidean-distance distribution and the constant power property of the 4D modulation. The geometric shaping of the proposed scheme is easy to implement and is robust to Mach-Zehnder modulator (MZM) imbalances and quantization errors stemming from the finite digital-to-analog converter (DAC) resolution. This robustness is compared to the one of other geometric-shaped non-linearity tolerant 4D schemes such as the 4D-2A-8PSK and the 4D-64PRS that can be both outperformed by our scheme in severe conditions.
• Novel Distribution Matcher Design for Short Length Frames Based on Non-Binary Convolutional Codes
  
  • Klaimi Rami
  • Abouseif Akram
  • Rekaya Ghaya
  • Jaouën Yves
  , 2023. We propose a distribution matcher that enables probabilistic constellation shaping while ensuring low-complexity dematching techniques. The proposal is based on non-binary convolutional codes, designed to respect a given optimal symbol distribution. In addition to lowering the dematching complexity, the proposed structure is shown to reduce the latency, to respect the target distribution with a low overhead and to outperform existing solutions with more than 0.3dB. It is also shown that, while being able to respect the target distribution for short frame lengths, the proposed technique helps enhancing the resilience of the optical system in question to the non-linearity effects.
• Multi-dimensional Energy Limitation in Sphere Shaping for Nonlinear Interference Noise Mitigation
  
  • Liu Jingtian
  • Awwad Élie
  • Jaouën Yves
  , 2023. We propose Four-Dimensional (4D) energy limit enumerative sphere shaping (ESS) of M-QAM signaling to minimize rate loss and improve the transmission performance over non-linear WDM optical-fiber systems. Simulation results show that the proposed scheme outperforms the conventional ESS by 0.19 bit/4D-symbol in achievable information rate over a 205-km single-span link and a WDM transmission of five polarization-division-multiplexed channels with 400-Gbit/s net rate per channel. We also study the achieved performance over several shaping block lengths and show that the achieved gains do not scale well over multi-span systems.
• Effective Impact of Modulation on Interchannel Nonlinear Effects in Realistic Submarine Links with Commercial Transceivers
  
  • Girard-Jollet Joana
  • Antona Jean-Christophe
  • Meseguer Alcatel Alexis Carbo
  • Dupont Sebastien
  • Garuz Richard
  • Zambrano Andrea Quintana
  • Othman Ghaya Rekaya-Ben
  , 2023, pp.1-5. We measure the GSNR of a subsea link with three different realtime transceivers for various link reaches and input powers. We experimentally show that, contrary to what the eGN model predicts, noise loading does not cause an additional penalty in the channel under test performance. (10.1109/ACP/POEM59049.2023.10369619)
  DOI : 10.1109/ACP/POEM59049.2023.10369619
• Exploring the potentials of online machine learning for predictive maintenance : a case study in the railway industry
  
  • Le Nguyen Minh-Huong
  • Turgis Fabien
  • Fayemi Pierre-Emmanuel
  • Bifet Albert
  Applied Intelligence, Springer Verlag, 2023, 53, pp.29758–29780. This study addresses data-driven predictive maintenance, an area in which machine learning has received considerable attention. Traditionally, a machine learning model is trained on static data before being put into production to predict failures on incoming data. However, new data typically present novelties that were not included in the training data, such as unexpected anomalies or faults. Such novelties reduce the model accuracy and require model retraining, which we consider to be a suboptimal practice.Therefore, we propose to leverage online machine learning as an adaptive and continuous alternative to implement efficient predictive maintenance on systems that produce data continuously. The literature on predictive maintenance concentrates primarily on failure prediction, whereas there are multiple stages in a standard predictive maintenance framework, such as data preprocessing and diagnostics, that require attention. In this study, we propose a modular pipeline consisting of three modules to execute many stages inside a predictive maintenance solution. Each module represents one of our original contributions. Firstly, because a system generates repeating patterns in the form of cycles when performing its functions, we construct an online active learning-based framework to extract these cycles from a stream of sensor data (cycle extraction with InterCE). Secondly, we implement an autoencoder for encoding the extracted cycles into feature vectors (feature learning with LSTM-AE). Thirdly, we develop an adaptive scoring function to compute the health of any system at any time using online clustering on the stream of feature vectors (health detection with CheMoc). These three contributions establish our framework for processing raw sensor data for predictive maintenance. We evaluate our methods using a real-world data set provided by SNCF, the French national railway company. For each experiment, we simulate a data stream consisting of sequentially arriving data from the provided data set to test our online algorithms. The experimental results demonstrate that (i) InterCE is able to extract cycles from a high-speed stream with greater accuracy than a hand-crafted expert system, (ii) LSTM-AE can identify meaningful features from the extracted cycles, and (iii) CheMoc can discover clusters that represent physical anomalies of the systems and capture the health evolution of the monitored systems. Due to a lack of ground-truth data at the time of writing, we have not implemented the prognostics method and will reserve this for future works. This study confirms the potential of online machine learning as an adaptive and lifelong learning solution for predictive maintenance. (10.1007/s10489-023-05092-4)
  DOI : 10.1007/s10489-023-05092-4
• Patch-Based Stochastic Attention for Image Editing
  
  • Cherel Nicolas
  • Almansa Andrés
  • Gousseau Yann
  • Newson Alasdair
  Computer Vision and Image Understanding, Elsevier, 2023, 238, pp.103866. Attention mechanisms have become of crucial importance in deep learning in recent years. These non-local operations, which are similar to traditional patch-based methods in image processing, complement local convolutions. However, computing the full attention matrix is an expensive step with heavy memory and computational loads. These limitations curb network architectures and performances, in particular for the case of high resolution images. We propose an efficient attention layer based on the stochastic algorithm PatchMatch, which is used for determining approximate nearest neighbors. We refer to our proposed layer as a “Patch-based Stochastic Attention Layer” (PSAL). Furthermore, we propose different approaches, based on patch aggregation, to ensure the differentiability of PSAL, thus allowing end-to-end training of any network containing our layer. PSAL has a small memory footprint and can therefore scale to high resolution images. It maintains this footprint without sacrificing spatial precision and globality of the nearest neighbors, which means that it can be easily inserted in any level of a deep architecture, even in shallower levels. We demonstrate the usefulness of PSAL on several image editing tasks, such as image inpainting, guided image colorization, and single-image super-resolution. Our code is available at: https://github.com/ncherel/psal (10.1016/j.cviu.2023.103866)
  DOI : 10.1016/j.cviu.2023.103866
• Pulmonary emphysema subtypes defined by unsupervised machine learning on CT scans
  
  • Angelini Elsa
  • Yang Jie
  • Balte Pallavi
  • Hoffman Eric
  • Manichaikul Ani
  • Sun Yifei
  • Shen Wei
  • Austin John
  • Allen Norrina
  • Bleecker Eugene
  • Bowler Russell
  • Cho Michael
  • Cooper Christopher
  • Couper David
  • Dransfield Mark
  • Garcia Christine Kim
  • Han Meilan
  • Hansel Nadia
  • Hughes Emlyn
  • Jacobs David
  • Kasela Silva
  • Kaufman Joel Daniel
  • Kim John Shinn
  • Lappalainen Tuuli
  • Lima Joao
  • Malinsky Daniel
  • Martinez Fernando
  • Oelsner Elizabeth
  • Ortega Victor
  • Paine Robert
  • Post Wendy
  • Pottinger Tess
  • Prince Martin
  • Rich Stephen
  • Silverman Edwin
  • Smith Benjamin
  • Swift Andrew
  • Watson Karol
  • Woodruff Prescott
  • Laine Andrew
  • Barr R Graham
  Thorax, BMJ Publishing Group, 2023, 78 (11), pp.1067-1079. Background Treatment and preventative advances for chronic obstructive pulmonary disease (COPD) have been slow due, in part, to limited subphenotypes. We tested if unsupervised machine learning on CT images would discover CT emphysema subtypes with distinct characteristics, prognoses and genetic associations. Methods New CT emphysema subtypes were identified by unsupervised machine learning on only the texture and location of emphysematous regions on CT scans from 2853 participants in the Subpopulations and Intermediate Outcome Measures in COPD Study (SPIROMICS), a COPD case–control study, followed by data reduction. Subtypes were compared with symptoms and physiology among 2949 participants in the population-based Multi-Ethnic Study of Atherosclerosis (MESA) Lung Study and with prognosis among 6658 MESA participants. Associations with genome-wide single-nucleotide-polymorphisms were examined. Results The algorithm discovered six reproducible (interlearner intraclass correlation coefficient, 0.91–1.00) CT emphysema subtypes. The most common subtype in SPIROMICS, the combined bronchitis-apical subtype, was associated with chronic bronchitis, accelerated lung function decline, hospitalisations, deaths, incident airflow limitation and a gene variant near DRD1 , which is implicated in mucin hypersecretion (p=1.1 ×10 −8 ). The second, the diffuse subtype was associated with lower weight, respiratory hospitalisations and deaths, and incident airflow limitation. The third was associated with age only. The fourth and fifth visually resembled combined pulmonary fibrosis emphysema and had distinct symptoms, physiology, prognosis and genetic associations. The sixth visually resembled vanishing lung syndrome. Conclusion Large-scale unsupervised machine learning on CT scans defined six reproducible, familiar CT emphysema subtypes that suggest paths to specific diagnosis and personalised therapies in COPD and pre-COPD. (10.1136/thorax-2022-219158)
  DOI : 10.1136/thorax-2022-219158
• The Glass Ceiling of Automatic Evaluation in Natural Language Generation
  
  • Colombo Pierre
  • Peyrard Maxime
  • Noiry Nathan
  • West Robert
  • Piantanida Pablo
  , 2022, pp.178-183. Automatic evaluation metrics capable of replacing human judgments are critical to allowing fast development of new methods. Thus, numerous research efforts have focused on crafting such metrics. In this work, we take a step back and analyze recent progress by comparing the body of existing automatic metrics and human metrics altogether. As metrics are used based on how they rank systems, we compare metrics in the space of system rankings. Our extensive statistical analysis reveals surprising findings: automatic metrics – old and new – are much more similar to each other than to humans. Automatic metrics are not complementary and rank systems similarly. Strikingly, human metrics predict each other much better than the combination of all automatic metrics used to predict a human metric. It is surprising because human metrics are often designed to be independent, to capture different aspects of quality, e.g. content fidelity or readability. We provide a discussion of these findings and recommendations for future work in the field of evaluation. (10.18653/v1/2023.findings-ijcnlp.16)
  DOI : 10.18653/v1/2023.findings-ijcnlp.16
• Singer Identity Representation Learning using Self-Supervised Techniques
  
  • Torres Bernardo
  • Lattner Stefan
  • Richard Gael
  , 2023. Significant strides have been made in creating voice identity representations using speech data. However, the same level of progress has not been achieved for singing voices. To bridge this gap, we suggest a framework for training singer identity encoders to extract representations suitable for various singing-related tasks, such as singing voice similarity and synthesis. We explore different selfsupervised learning techniques on a large collection of isolated vocal tracks and apply data augmentations during training to ensure that the representations are invariant to pitch and content variations. We evaluate the quality of the resulting representations on singer similarity and identification tasks across multiple datasets, with a particular emphasis on out-of-domain generalization. Our proposed framework produces high-quality embeddings that outperform both speaker verification and wav2vec 2.0 pre-trained baselines on singing voice while operating at 44.1 kHz. We release our code and trained models to facilitate further research on singing voice and related areas.
• Concentration bounds for the empirical angular measure with statistical learning applications
  
  • Clémençon Stéphan
  • Jalalzai Hamid
  • Lhaut Stéphane
  • Sabourin Anne
  • Segers Johan
  Bernoulli, Bernoulli Society for Mathematical Statistics and Probability, 2023, 29 (4). The angular measure on the unit sphere characterizes the first-order dependence structure of the components of a random vector in extreme regions and is defined in terms of standardized margins. Its statistical recovery is an important step in learning problems involving observations far away from the center. In the common situation that the components of the vector have different distributions, the rank transformation offers a convenient and robust way of standardizing data in order to build an empirical version of the angular measure based on the most extreme observations. However, the study of the sampling distribution of the resulting empirical angular measure is challenging. It is the purpose of the paper to establish finite-sample bounds for the maximal deviations between the empirical and true angular measures, uniformly over classes of Borel sets of controlled combinatorial complexity. The bounds are valid with high probability and, up to logarithmic factors, scale as the square root of the effective sample size. The bounds are applied to provide performance guarantees for two statistical learning procedures tailored to extreme regions of the input space and built upon the empirical angular measure: binary classification in extreme regions through empirical risk minimization and unsupervised anomaly detection through minimumvolume sets of the sphere. (10.3150/22-BEJ1562)
  DOI : 10.3150/22-BEJ1562
• A Novel Information-Theoretic Objective to Disentangle Representations for Fair Classification
  
  • Colombo Pierre
  • Noiry Nathan
  • Staerman Guillaume
  • Piantanida Pablo
  , 2023, pp.184-198. One of the pursued objectives of deep learning is to provide tools that learn abstract representations of reality from the observation of multiple contextual situations. More precisely, one wishes to extract disentangled representations which are (i) low dimensional and (ii) whose components are independent and correspond to concepts capturing the essence of the objects under consideration (Locatello et al., 2019b). One step towards this ambitious project consists in learning disentangled representations with respect to a predefined (sensitive) attribute, e.g., the gender or age of the writer. Perhaps one of the main application for such disentangled representations is fair classification. Existing methods extract the last layer of a neural network trained with a loss that is composed of a cross-entropy objective and a disentanglement regularizer. In this work, we adopt an information-theoretic view of this problem which motivates a novel family of regularizers that minimizes the mutual information between the latent representation and the sensitive attribute conditional to the target. The resulting set of losses, called CLINIC, is parameter free and thus, it is easier and faster to train. CLINIC losses are studied through extensive numerical experiments by training over 2k neural networks. We demonstrate that our methods offer a better disentanglement/accuracy trade-off than previous techniques, and generalize better than training with cross-entropy loss solely provided that the disentanglement task is not too constraining. (10.18653/v1/2023.findings-ijcnlp.17)
  DOI : 10.18653/v1/2023.findings-ijcnlp.17
• STUDD: a student-teacher method for unsupervised concept drift detection
  
  • Cerqueira Vítor
  • Gomes Heitor Murilo
  • Bifet Albert
  • Torgo Luís
  Machine Learning, Springer Verlag, 2023, 112 (11), pp.4351--4378. Concept drift detection is a crucial task in data stream evolving environments. Most of state of the art approaches designed to tackle this problem monitor the loss of predictive models. However, this approach falls short in many real-world scenarios, where the true labels are not readily available to compute the loss. In this context, there is increasing attention to approaches that perform concept drift detection in an unsupervised manner, i.e., without access to the true labels after the model is deployed. We propose a novel approach to unsupervised concept drift detection based on a student-teacher learning paradigm. Essentially, we create an auxiliary model (student) to mimic the primary model’s behaviour (teacher). At run-time, our approach is to use the teacher for predicting new instances and monitoring the mimicking loss of the student for concept drift detection. In a set of experiments using 19 data streams, we show that the proposed approach can detect concept drift and present a competitive behaviour relative to the state of the art approaches. (10.1007/S10994-022-06188-7)
  DOI : 10.1007/S10994-022-06188-7
• 5G vs Wifi6 downlink power consumption comparison for teleworking use case
  
  • Hentati Mohamed Yassine
  • Chahed Tijani
  • Ciblat Philippe
  • Coupechoux Marceau
  • Najeh Sameh
  , 2023, pp.1-9. eleworking became a vastly popular practice lately owing to the Covid-19 outbreak. Our present study aims to evaluate the operating energy consumption in 5G versus Wifi6 networks for this remote working case study. We specifically study the detailed operations of transmission and reception at both radio access networks and quantify their power consumption in the downlink, analytically, using simulations and, for the case of Wifi6. We focus on three case studies in the framework of a teleworking tool: audio session, video session and shared screen session. Our results show that for the use case of video session in the downlink, when the number of users is rather small, Wifi6 consumes less power, however, as the number of users gets larger, 5G with Multi-User MIMO (MU-MIMO) outperforms the individual Wifi accesses. These results depend on the number of simultaneous MUMIMO streams as well as on the fixed component of the power consumption of the 5G, for which we present a sensitivity analysis as well. (10.1109/ComNet60156.2023.10366633)
  DOI : 10.1109/ComNet60156.2023.10366633
• THE HI-AUDIO ONLINE PLATFORM FOR DISTRIBUTED MUSIC CROWDSOURCING DATABASE COLLECTION
  
  • Gil Panal Jose Manuel
  • David Aurélien
  • Richard Gael
  , 2023. We present in this paper the recent development of an online platform for musicians, researchers and an open community of enthusiasts of audio and music with a view to build a public database of music recordings from a wide variety of styles and different cultures. The data generated and collected will primarily be audio data, coming from various sources, including field recordings, existing datasets, and users’ collaboration. The platform aims at gathering a distributed music crowdsourcing database collection where each music piece is built from asynchronous recordings of different tracks at remote sites. The complete tool and databases generated will be openly distributed for research purposes.
• A Statistical Learning View of Simple Kriging
  
  • Siviero Emilia
  • Chautru Emilie
  • Clémençon Stéphan
  Test, Spanish Society of Statistics and Operations Research/Springer, 2023. In the Big Data era, with the ubiquity of geolocation sensors in particular, massive datasets exhibiting a possibly complex spatial dependence structure are becoming increasingly available. In this context, the standard probabilistic theory of statistical learning does not apply directly and guarantees of the generalization capacity of predictive rules learned from such data are left to establish. We analyze here the simple Kriging task, the flagship problem in Geostatistics: the values of a square integrable random field $X=\{X_s\}_{s\in S}$, $S\subset \mathbb{R}^2$, with unknown covariance structure are to be predicted with minimum quadratic risk, based upon observing a single realization of the spatial process at a finite number of locations $s_1,\; \ldots,\; s_n$ in $S$. Despite the connection of this minimization problem with kernel ridge regression, establishing the generalization capacity of empirical risk minimizers is far from straightforward, due to the non i.i.d. nature of the spatial data $X_{s_1},\; \ldots,\; X_{s_n}$ involved. In this article, nonasymptotic bounds of order $O_{\mathbb{P}}(1/n)$ are proved for the excess risk of a plug-in predictive rule mimicking the true minimizer in the case of isotropic stationary Gaussian processes observed at locations forming a regular grid. These theoretical results, as well as the role played by the technical conditions required to establish them, are illustrated by various numerical experiments and hopefully pave the way for further developments in statistical learning based on spatial data.
• Key Filtering in Cube Attacks from the Implementation Aspect
  
  • Fan Hao
  • Hao Yonglin
  • Wang Qingju
  • Gong Xinxin
  • Jiao Lin
  , 2023, 14342, pp.293-317. In cube attacks, key filtering is a basic step of identifying the correct key candidates by referring to the truth tables of superpolies. When terms of superpolies get massive, the truth table lookup complexity of key filtering increases significantly. In this paper, we propose the concept of implementation dependency dividing all cube attacks into two categories: implementation dependent and implementation independent. The implementation dependent cube attacks can only be feasible when the assumption that one encryption oracle query is more complicated than one table lookup holds. On the contrary, implementation independent cube attacks remain feasible in the extreme case where encryption oracles are implemented in the full codebook manner making one encryption query equivalent to one table lookup. From this point of view, we scrutinize existing cube attack results of stream ciphers TRIVIUM, Grain-128AEAD, ACORN and KREYVIUM. As a result, many of them turn out to be implementation dependent. Combining with the degree evaluation and divide-and-conquer techniques used for superpoly recovery, we further propose new cube attack results on KREYVIUM reduced to 898, 899 and 900 rounds. Such new results not only mount to the maximal number of rounds so far but also are implementation independent. (10.1007/978-981-99-7563-1_14)
  DOI : 10.1007/978-981-99-7563-1_14
• Privacy-Preserving Digital Vaccine Passport
  
  • Duong Thai
  • Gao Jiahui
  • Phan Duong Hieu
  • Trieu Ni
  , 2023, 14342, pp.137-161. The global lockdown imposed during the Covid-19 pandemic has resulted in significant social and economic challenges. In an effort to reopen economies and simultaneously control the spread of the disease, the implementation of contact tracing and digital vaccine passport technologies has been introduced. While contact tracing methods have been extensively studied and scrutinized for security concerns through numerous publications, vaccine passports have not received the same level of attention in terms of defining the problems they address, establishing security requirements, or developing efficient systems. Many of the existing methods employed currently suffer from privacy issues. This work introduces PPass, an advanced digital vaccine passport system that prioritizes user privacy. We begin by outlining the essential security requirements for an ideal vaccine passport system. To address these requirements, we present two efficient constructions that enable PPass to function effectively across various environments while upholding user privacy. By estimating its performance, we demonstrate the practical feasibility of PPass. Our findings suggest that PPass can efficiently verify a passenger's vaccine passport in just 7 milliseconds, with a modest bandwidth requirement of 480KB. (10.1007/978-981-99-7563-1_7)
  DOI : 10.1007/978-981-99-7563-1_7
• An Introduction to ALISA and Its Usage for an Industrial Railway System Case Study
  
  • Blouin Dominique
  • Crisafulli Paolo
  • Maxim Cristian
  • Caron Francoise
  Ada Letters, Association for Computing Machinery, 2023, 43 (1), pp.69-72. This paper presents an overview of ALISA (Architecture- Led Incremental System Assurance) and its evaluation for a case study of the railway domain as presented during the ADEPT workshop collocated with the 26th Ada-Europe International Conference on Reliable Software Technologies. (10.1145/3631483.3631493)
  DOI : 10.1145/3631483.3631493
• Dually Computable Cryptographic Accumulators and Their Application to Attribute Based Encryption
  
  • Barthoulot Anaïs
  • Blazy Olivier
  • Canard Sébastien
  , 2023, 14342, pp.538-562. In 1993, Benaloh and De Mare introduced cryptographic accumulator, a primitive that allows the representation of a set of values by a short object (the accumulator) and offers the possibility to prove that some input values are in the accumulator. For this purpose, so-called asymmetric accumulators require the creation of an additional cryptographic object, called a witness. Through the years, several instantiations of accumulators were proposed either based on number theoretic assumptions, hash functions, bilinear pairings or more recently lattices. In this work, we present the first instantiation of an asymmetric cryptographic accumulator that allows private computation of the accumulator but public witness creation. This is obtained thanks to our unique combination of the pairing based accumulator of Nguyen with dual pairing vector spaces. We moreover introduce the new concept of dually computable cryptographic accumulators, in which we offer two ways to compute the representation of a set: either privately (using a dedicated secret key) or publicly (using only the scheme's public key), while there is a unique witness creation for both cases. All our constructions of accumulators have constant size accumulated value and witness, and satisfy the accumulator security property of collision resistance, meaning that it is not possible to forge a witness for an element that is not in the accumulated set. As a second contribution, we show how our new concept of dually computable cryptographic accumulator can be used to build a Ciphertext Policy Attribute Based Encryption (CP-ABE). Our resulting scheme permits policies expressed as disjunctions of conjunctions (without "NO" gates), and is adaptively secure in the standard model. This is the first CP-ABE scheme having both constant-size user secret keys and ciphertexts (i.e. independent of the number of attributes in the scheme, or the policy size). For the first time, we provide a way to use cryptographic accumulators for both key management and encryption process. (10.1007/978-981-99-7563-1_24)
  DOI : 10.1007/978-981-99-7563-1_24
• NOMA-based Scheduling and offloading for energy harvesting devices using reinforcement learning
  
  • Djemai Ibrahim
  • Sarkiss Mireille
  • Ciblat Philippe
  , 2024. We consider a joint optimization problem of resource scheduling and computation offloading in a Mobile-Edge Computing (MEC) system where User Equipments (UEs) or devices have energy harvesting functionalities. The UEs can either execute locally the data packets or offload them to a nearby MEC server for remote processing. The main objective is to minimize the overall packet losses of the UEs under strict delay constraints imposed by applications. Non-Orthogonal Multiple Access is enabled to allow UEs sending their data packets simultaneously. The problem is formulated as a Markov Decision Process and is solved using Proximal Policy Optimization, a Deep Reinforcement Learning algorithm. The numerical results show the efficiency of such an algorithm in reducing the packet loss as well as the energy consumed during testing compared to some naive heuristics (10.1109/IEEECONF59524.2023.10476942)
  DOI : 10.1109/IEEECONF59524.2023.10476942
• A gem5 based Platform for Micro-Architectural Security Analysis
  
  • Forcioli Quentin
  • Danger Jean-Luc
  • Chaudhuri Sumanta
  , 2023, pp.91-99. In this article we present a simulation platform based on gem5 for security analysis. On top of gem5’s architectural exploration and performance estimation capability, our platform permits attacks on ARM Trustzone, security evaluation of cypto libraries, and attacks from accelerators or 3rd party IPs present in the SoC. We discuss various components of our platform such as GDB, gem5, SystemC TLM 2.0 and the steps to boot an open source trusted execution environment called OPTEE. We present an in-vitro experimental attack in Syscall mode on the mbedTLS library and we show how this attack can be fine-tuned. We also present two in-vivo attacks on OPTEE on the RSA signing Trustlet and the Secure Storage Trustlet to demonstrate the capabilities and usage of our platform (10.1145/3623652.3623674)
  DOI : 10.1145/3623652.3623674
• teex: A toolbox for the evaluation of explanations
  
  • Antonanzas Jesus
  • Jia Yunzhe
  • Frank Eibe
  • Bifet Albert
  • Pfahringer Bernhard
  Neurocomputing, Elsevier, 2023, 555, pp.126642. We present teex, a Python toolbox for the evaluation of explanations. teex focuses on the evaluation of local explanations of the predictions of machine learning models by comparing them to ground-truth explanations. It supports several types of explanations: feature importance vectors, saliency maps, decision rules, and word importance maps. A collection of evaluation metrics is provided for each type. Real-world datasets and generators of synthetic data with ground-truth explanations are also contained within the library. teex contributes to research on explainable AI by providing tested, streamlined, user-friendly tools to compute quality metrics for the evaluation of explanation methods. Source code and a basic overview can be found at github.com/chus-chus/teex, and tutorials and full API documentation are at teex.readthedocs.io. (10.1016/J.NEUCOM.2023.126642)
  DOI : 10.1016/J.NEUCOM.2023.126642
• Enabling programmable deterministic communications in 6G
  
  • Thi Minh-Thuyen
  • Ben Hadj Said Siwar
  • Roberty Adrien
  • Chbib Fadlallah
  • Khatoun Rida
  • Linguaglossa Leonardo
  , 2023. Emerging applications and technologies such as vehicle-to-everything (V2X), edge-computing, and artificial intelligence have emphasized the demand for low-latency and deterministic communication. Although the 5G network has taken several efforts to fulfill this demand, such as with 5G-Time-Sensitive Networking (TSN) integration and DetNet, these efforts must be significantly expanded in 6G to fully achieve end-to-end deterministic communication. In this paper, we explore the problem of programmable deterministic communication in the new architecture of 6G. To deal with this problem, we rely on TSN, which has been proven to be a promising solution for deterministic communication. We take V2X as a use case, then investigate the two greatest challenges of this use case: low-latency communication and programmable network management for deterministic communication. To deal with these challenges, we introduce two solutions: (i) TSN low-latency scheduling supported by Multi-Agent Deep Reinforcement Learning, and (ii) programmable network management supported by SDN and joint cloud-infrastructure control. For each solution, detailed architecture and functionality design are presented. We show their high feasibility, applicability, and potentialities through comprehensive definitions, detailed explanations and in-depth qualitative analysis.
• Time-Domain Audio Source Separation Based on Gaussian Processes with Deep Kernel Learning
  
  • Nugraha Aditya Arie
  • Carlo Diego Di
  • Bando Yoshiaki
  • Fontaine Mathieu
  • Yoshii Kazuyoshi
  , 2023. This paper revisits single-channel audio source separation based on a probabilistic generative model of a mixture signal defined in the continuous time domain. We assume that each source signal follows a non-stationary Gaussian process (GP), i.e., any finite set of sampled points follows a zero-mean multivariate Gaussian distribution whose covariance matrix is governed by a kernel function over time-varying latent variables. The mixture signal composed of such source signals thus follows a GP whose covariance matrix is given by the sum of the source covariance matrices. To estimate the latent variables from the mixture signal, we use a deep neural network with an encoder-separator-decoder architecture (e.g., Conv-TasNet) that separates the latent variables in a pseudo-time-frequency space. The key feature of our method is to feed the latent variables into the kernel function for estimating the source covariance matrices, instead of using the decoder for directly estimating the time-domain source signals. This enables the decomposition of a mixture signal into the source signals with a classical yet powerful Wiener filter that considers the full covariance structure over all samples. The kernel function and the network are trained jointly in the maximum likelihood framework. Comparative experiments using two-speech mixtures under clean, noisy, and noisy-reverberant conditions from the WSJ0-2mix, WHAM!, and WHAMR! benchmark datasets demonstrated that the proposed method performed well and outperformed the baseline method under noisy and noisy-reverberant conditions.
• On the Trade-off between Over-smoothing and Over-squashing in Deep Graph Neural Networks
  
  • Giraldo Jhony H
  • Skianis Konstantinos
  • Bouwmans Thierry
  • Malliaros Fragkiskos D.
  , 2023, pp.566-576. Graph Neural Networks (GNNs) have succeeded in various computer science applications, yet deep GNNs underperform their shallow counterparts despite deep learning's success in other domains. Over-smoothing and over-squashing are key challenges when stacking graph convolutional layers, hindering deep representation learning and information propagation from distant nodes. Our work reveals that over-smoothing and over-squashing are intrinsically related to the spectral gap of the graph Laplacian, resulting in an inevitable trade-off between these two issues, as they cannot be alleviated simultaneously. To achieve a suitable compromise, we propose adding and removing edges as a viable approach. We introduce the Stochastic Jost and Liu Curvature Rewiring (SJLR) algorithm, which is computationally efficient and preserves fundamental properties compared to previous curvature-based methods. Unlike existing approaches, SJLR performs edge addition and removal during GNN training while maintaining the graph unchanged during testing. Comprehensive comparisons demonstrate SJLR's competitive performance in addressing over-smoothing and over-squashing. CCS CONCEPTS • Computing methodologies → Machine learning algorithms; • Computer systems organization → Neural networks. (10.1145/3583780.3614997)
  DOI : 10.1145/3583780.3614997