Publications

2024

• Any2Graph: Deep End-To-End Supervised Graph Prediction With An Optimal Transport Loss
  
  • Krzakala Paul
  • Yang Junjie
  • Flamary Rémi
  • d'Alché-Buc Florence
  • Laclau Charlotte
  • Labeau Matthieu
  , 2024. We propose Any2graph, a generic framework for end-to-end Supervised Graph Prediction (SGP) i.e. a deep learning model that predicts an entire graph for any kind of input. The framework is built on a novel Optimal Transport loss, the Partially-Masked Fused Gromov-Wasserstein, that exhibits all necessary properties (permutation invariance, differentiability and scalability) and is designed to handle any-sized graphs. Numerical experiments showcase the versatility of the approach that outperform existing competitors on a novel challenging synthetic dataset and a variety of real-world tasks such as map construction from satellite image (Sat2Graph) or molecule prediction from fingerprint (Fingerprint2Graph). (10.48550/arXiv.2402.12269)
  DOI : 10.48550/arXiv.2402.12269
• Continuous Product Graph Neural Networks
  
  • Einizade Aref
  • Malliaros Fragkiskos D.
  • Giraldo Jhony H.
  , 2024. Processing multidomain data defined on multiple graphs holds significant potential in various practical applications in computer science. However, current methods are mostly limited to discrete graph filtering operations. Tensorial partial differential equations on graphs (TPDEGs) provide a principled framework for modeling structured data across multiple interacting graphs, addressing the limitations of the existing discrete methodologies. In this paper, we introduce Continuous Product Graph Neural Networks (CITRUS) that emerge as a natural solution to the TPDEG. CITRUS leverages the separability of continuous heat kernels from Cartesian graph products to efficiently implement graph spectral decomposition. We conduct thorough theoretical analyses of the stability and over-smoothing properties of CITRUS in response to domain-specific graph perturbations and graph spectra effects on the performance. We evaluate CITRUS on well-known traffic and weather spatiotemporal forecasting datasets, demonstrating superior performance over existing approaches. The implementation codes are available at https://github.com/ArefEinizade2/CITRUS.
• TabEBM: A Tabular Data Augmentation Method with Distinct Class-Specific Energy-Based Models
  
  • Margeloiu Andrei
  • Jiang Xiangjian
  • Simidjievski Nikola
  • Jamnik Mateja
  , 2024, pp.72094-72144. <div xmlns="http://www.tei-c.org/ns/1.0"><p>Data collection is often difficult in critical fields such as medicine, physics, and chemistry, yielding typically only small tabular datasets. However, classification methods tend to struggle with these small datasets, leading to poor predictive performance. Increasing the training set with additional synthetic data, similar to data augmentation in images, is commonly believed to improve downstream tabular classification performance. However, current tabular generative methods that learn either the joint distribution p(x, y) or the class-conditional distribution p(x | y) often overfit on small datasets, resulting in poor-quality synthetic data, usually worsening classification performance compared to using real data alone. To solve these challenges, we introduce TabEBM, a novel class-conditional generative method using Energy-Based Models (EBMs). Unlike existing tabular methods that use a shared model to approximate all class-conditional densities, our key innovation is to create distinct EBM generative models for each class, each modelling its class-specific data distribution individually. This approach creates robust energy landscapes, even in ambiguous class distributions. Our experiments show that TabEBM generates synthetic data with higher quality and better statistical fidelity than existing methods. When used for data augmentation, our synthetic data consistently leads to improved classification performance across diverse datasets of various sizes, especially small ones. Code is available at <ref type="url" target="https://github.com/andreimargeloiu/TabEBM">https://github.com/andreimargeloiu/TabEBM</ref>.</p></div> (10.52202/079017-2302)
  DOI : 10.52202/079017-2302
• Leveraging Sentiment and Emotion Analysis to Enhance Cyberbullying Detection
  
  • Berjawi Omran
  • Khatoun Rida
  • Fahs Walid
  • Fenza Giuseppe
  , 2024. Cyberbullying has emerged as a critical concern in the age of social media, where anonymity and widespread access facilitate abusive behaviors. This paper explores the effectiveness of advanced machine learning techniques combined with sentiment and emotion analysis for cyberbullying detection. We utilized a dataset of tweets and evaluated various models, including Logistic Regression, Support Vector Machine, Random Forest, and XGBoost, to identify the most effective approaches. Our proposed model, which integrates TF-IDF with sentiment and emotion scores, achieved a high accuracy of 0.9890, outperforming established models such as those based on Random Forest with GloVe and advanced methods like RoBERTa with GloVe and PCA. Our analysis further revealed distinct emotional patterns associated with different categories of cyberbullying, with negative emotions such as anger, disgust, and fear being predominantly linked to cyberbullying content. In contrast, non-cyberbullying content displayed a more balanced emotional profile, exhibiting higher values for neutral and positive emotions. These findings underscore the significant role of emotional and sentiment analysis in enhancing the detection of harmful behaviors in online environments. (10.1109/SNAMS64316.2024.10883815)
  DOI : 10.1109/SNAMS64316.2024.10883815
• Methods and tools for probabilistic analytical reliability analysis of gate netlists
  
  • Goudet Esther
  , 2024. The analysis of complex SoCs for ISO26262 automotive certification requires the extraction of reliability metrics such as the logic masking rate of a circuit, which is used to calculate fault propagation probabilities. Such metrics can be extracted using experimental methods, such as fault injection, or through analytical methods. This thesis proposes advanced probabilistic methods to compute various reliability metrics from a circuit’s netlist. These metrics will be used in fault models and reliability assessment processes such as RAMS and MBSA approaches. During a previous PhD, the HCPM model, based on netlist partitioning, was patented. This thesis builds upon the results of that PhD, extends and completes the developments of the HCPM model, and draws on the partitioning method to design a new analysis model for combinational circuits, capable of handling a wider range of circuits. In the developed method, the analysis of the output signal error rates is carried out separately in each cluster, using the same probabilistic model for all blocks in the partition. To determine the error rate of a combinational circuit, the output error rates of the clusters are transmitted from block to block, until reaching the primary output signals of the circuit. The analysis complexity of the entire circuit is thus reduced to the complexity of the heaviest cluster. Under specific partitioning constraints, the overall study of the circuit remains accurate, while the time and memory required for the analysis are significantly curtailed. Throughout the thesis, the reliability results obtained are compared with those from other state-of-the-art analytical models and cross-checked by comparing them with the rates obtained using a fault injection tool implemented on an FPGA. The aspects of our model that are subject to reflection and debates are (1) the relevance of the distribution of the faults simulated by our algorithm in a circuit, and (2) the consideration of the probabilities of the states of reconverging signals in the error rate calculations. A reconvergent fan-out that spreads across several clusters leads to an incorrect estimation of the error rate of a combinational circuit. Metrics extraction will focus on feeding RAMS and MBSA frameworks to allow composition of the results to assess safety at the system level from results obtained at the block level using the developed method. A second development in the thesis focuses on more physically realistic methods, such as the Transistor-PTM model, based on the transistor schematic of a logic gate. Fault models for other elementary components of a silicon library will also be explored to link analytical reliability models to silicon SoC reliability.
• Annealed Multiple Choice Learning: Overcoming limitations of Winner-takes-all with annealing
  
  • Perera David
  • Letzelter Victor
  • Mariotte Théo
  • Cortés Adrien
  • Chen Mickael
  • Essid Slim
  • Richard Gaël
  , 2024. We introduce Annealed Multiple Choice Learning (aMCL) which combines simulated annealing with MCL. MCL is a learning framework handling ambiguous tasks by predicting a small set of plausible hypotheses. These hypotheses are trained using the Winner-takes-all (WTA) scheme, which promotes the diversity of the predictions. However, this scheme may converge toward an arbitrarily suboptimal local minimum, due to the greedy nature of WTA. We overcome this limitation using annealing, which enhances the exploration of the hypothesis space during training. We leverage insights from statistical physics and information theory to provide a detailed description of the model training trajectory. Additionally, we validate our algorithm by extensive experiments on synthetic datasets, on the standard UCI benchmark, and on speech separation.
• Adaptive Hardcore Bit and Quantum Key Leasing over Classical Channel from LWE with Polynomial Modulus
  
  • Phan Duong Hieu
  • Wen Weiqiang
  • Yan Xingyu
  • Zheng Jinwei
  , 2025, 15492, pp.185-214. Quantum key leasing, also known as public key encryption with secure key leasing (PKE-SKL), allows a user to lease a (quantum) secret key to a server for decryption purpose, with the capability of revoking the key afterwards. In the pioneering work by Chardouvelis et al. (arXiv:2310.14328), a PKE-SKL scheme utilizing classical channels was successfully built upon the noisy trapdoor claw-free (NTCF) family. This approach, however, relies on the superpolynomial hardness of learning with errors (LWE) problem, which could affect both efficiency and security of the scheme. In our work, we demonstrate that the reliance on superpolynomial hardness is unnecessary, and that LWE with polynomial-size modulus is sufficient to achieve the same goal. Our approach enhances both efficiency and security, thereby improving the practical feasibility of the scheme on near-term quantum devices. To accomplish this, we first construct a noticeable NTCF (NNTCF) family with the adaptive hardcore bit property, based on LWE with polynomial-size modulus. To the best of our knowledge, this is the first demonstration of the adaptive hardcore bit property based on LWE with polynomial-size modulus, which may be of independent interest. Building on this foundation, we address additional challenges in prior work to construct the first PKE-SKL scheme satisfying the following properties: (i) the entire protocol utilizes only classical communication, and can also be lifted to support homomorphism. (ii) the security is solely based on LWE assumption with polynomial-size modulus. As a demonstration of the versatility of our noticeable NTCF, we show that an efficient proof of quantumness protocol can be built upon it. Specifically, our protocol enables a classical verifier to test the quantumness while relying exclusively on the LWE assumption with polynomial-size modulus. (10.1007/978-981-96-0947-5_7)
  DOI : 10.1007/978-981-96-0947-5_7
• Microarchitectural Timing Side-Channel Attacks
  
  • Khan Mahreen
  • Pacalet Renaud
  • Mushtaq Maria
  • Apvrille Ludovic
  , 2024. Microarchitectural timing side-channel attacks exploit variations in execution times caused by the underlying hardware to extract sensitive information. These attacks leverage architectural features like caches, branch predictors, and speculative execution.
• Multispectral Style Distances and Application to Texture Synthesis Using RGB Convolutional Neural Networks
  
  • Ollivier Sélim
  • Gousseau Yann
  • Lefebvre Sidonie
  , 2025, pp.1-5. State-of-the-art methods for RGB texture synthesis and style transfer leverage the representations learned by convolutional neural networks (CNN) on large datasets. Style distances, obtained by comparing statistics of deep features, play a pivotal role in synthesis procedures. Extending these distances to multispectral images is challenging because the pre-trained CNN only operate on RGB images. This work presents two multispectral style distances that still rely on a RGB CNN to avoid additional training. The first consists in a classical style distance, averaged over images formed by triplets of spectral bands. The second takes advantage of a projection of the multispectral pixels onto a three-dimensional space. We demonstrate their efficiency by performing multispectral texture synthesis. (10.1109/WHISPERS65427.2024.10876488)
  DOI : 10.1109/WHISPERS65427.2024.10876488
• Unsupervised Super-Resolution of Hyperspectral Remote Sensing Images Using Fully Synthetic Training
  
  • Xu Xinxin
  • Gousseau Yann
  • Kervazo Christophe
  • Ladjal Saïd
  , 2024, pp.1-5. Considerable work has been dedicated to hyperspectral single image super-resolution to improve the spatial resolution of hyperspectral images and fully exploit their potential. However, most of these methods are supervised and require some data with ground truth for training, which is often non-available. To overcome this problem, we propose a new unsupervised training strategy for the super-resolution of hyperspectral remote sensing images, based on the use of synthetic abundance data. Its first step decomposes the hyperspectral image into abundances and endmembers by unmixing. Then, an abundance super-resolution neural network is trained using synthetic abundances, which are generated using the dead leaves model in such a way as to faithfully mimic real abundance statistics. Next, the spatial resolution of the considered hyperspectral image abundances is increased using this trained network, and the high resolution hyperspectral image is finally obtained by recombination with the endmembers. Experimental results show the training potential of the synthetic images, and demonstrate the method effectiveness. (10.1109/WHISPERS65427.2024.10876452)
  DOI : 10.1109/WHISPERS65427.2024.10876452
• MinRank Gabidulin Encryption Scheme on Matrix Codes
  
  • Aragon Nicolas
  • Couvreur Alain
  • Dyseryn Victor
  • Gaborit Philippe
  • Vinçotte Adrien
  , 2024, LNCS-15487, pp.68-100. The McEliece scheme is a generic frame introduced in [28], which allows to use any error correcting code for which there exists an efficient decoding algorithm to design an encryption scheme by hiding the generator matrix of the code. Similarly, the Niederreiter frame, introduced in [30], is the dual version of the McEliece scheme, and achieves smaller ciphertexts. In the present paper, we propose a generalization of the McEliece and the Niederreiter frame to matrix codes and the MinRank problem, that we apply to Gabidulin matrix codes (Gabidulin rank codes considered as matrix codes). The masking we consider consists in starting from a rank code C, computing a matrix version of C and then concatenating a certain number of rows and columns to the matrix code version of the rank code C before applying an isometry for matrix codes, i.e. right and left multiplications by fixed random matrices. The security of the schemes relies on the MinRank problem to decrypt a ciphertext, and the structural security of the scheme relies on the new EGMC-Indistinguishability problem that we introduce and that we study in detail. The main structural attack that we propose consists in trying to recover the masked linearity over the extension field which is lost during the masking process. Overall, starting from Gabidulin codes, we obtain a very appealing trade off between the size of the ciphertext and the size of the public key. For 128 bits of security we propose parameters ranging from ciphertexts of size 65 B (and public keys of size 98 kB) to ciphertexts of size 138 B (and public keys of size 41 kB). For 256 bits of security, we can obtain ciphertext as low as 119 B, or public key as low as 87 kB. Our new approach permits to achieve a better trade-off between ciphertexts and public key than the classical McEliece scheme instantiated with Goppa codes. (10.1007/978-981-96-0894-2_3)
  DOI : 10.1007/978-981-96-0894-2_3
• Joint Optimization for Anti-jamming Communication with UAV-carried Intelligent Reflecting Surface
  
  • Yu Jinsong
  • Liu Lingya
  • Hua Cunqing
  • Gu Pengwenlong
  , 2024, pp.5120-5125. (10.1109/GLOBECOM52923.2024.10901788)
  DOI : 10.1109/GLOBECOM52923.2024.10901788
• ALICE: Adapt your Learnable Image Compression modEl for variable bitrates
  
  • Spadaro Gabriele
  • Ali Muhammad Salman
  • Presta Alberto
  • Pilo Giommaria
  • Bae Sung-Ho
  • Giraldo Jhony
  • Fiandrotti Attilio
  • Grangetto Marco
  • Tartaglione Enzo
  , 2024, pp.1-5. When training a Learned Image Compression model, the loss function is minimized such that the encoder and the decoder attain a target Rate-Distorsion trade-off. Therefore, a distinct model shall be trained and stored at the transmitter and receiver for each target rate, fostering the quest for efficient variable bitrate compression schemes. This paper proposes plugging Low-Rank Adapters into a transformer-based pre-trained LIC model and training them to meet different target rates. With our method, encoding an image at a variable rate is as simple as training the corresponding adapters and plugging them into the frozen pre-trained model. Our experiments show performance comparable with state-of-the-art fixed-rate LIC models at a fraction of the training and deployment cost. We publicly released the code at https://github.com/EIDOSLAB/ALICE. (10.1109/VCIP63160.2024.10849832)
  DOI : 10.1109/VCIP63160.2024.10849832
• Image Attribute Editing and Film Grain Rendering using Deep Neural Networks
  
  • Lesné Gwilherm
  , 2024. This thesis is divided in 6 chapters.The first is devoted to introducing the related topics. In particular, we present image editing and some of its tasks, giving background and motivations. We also highlight our contributions in the form of publications and presentations at colloquiums and conferences.The second chapter of this thesis describes and discusses the previous existing works on image generation and their applications. This wide topic includes image synthesis methods but is also linked to notions such as disentanglement and image editing in general. We beforehand present all paradigms to generate images using neural networks with their most famous variants. Then, disentanglement is defined and discussed. Finally, several editing methods are presented as they take advantage of the power of generative models.The third chapter focuses on our first contribution. That is, we propose a method to perform face image editing via displacements in the latent space of a style-type generative model. This kind of generative model appeared with StyleGAN and can be seen as generators having a latent space that does not follow an a priori distribution. In particular, we provide an approach to compress their latent space, keeping the minimum of necessary information, using PCA. Starting with this compressed space, we train a neural network to transform it into a new latent space where structure is imposed, in order to make it semantically interpretable. Namely, each axis of our new latent space is associated with a specific attribute of the image. This way, image editing can be performed easily by only modifying one dimension of our new latent space, drawing a parallel with the intuition of sliders controlling the semantics of the synthesized image.To demonstrate the relevance of our work, several evaluations are conducted, both qualitatively and quantitatively, and presented at the end of this chapter. Notably, we show that our approach can not only be used for attribute editing but also for constraining the semantics generated when solving inverse problems.The next part of this work (Chapter 4) includes a brief explanation of what is film grain and its physical phenomenon. Afterwards, we show and categorize several methods and models that aim at rendering this typical aspect of analog film. This chapter is completed by a review of texture synthesis and style transfer as our second contribution takes its inspiration from techniques developped in this field of research.In Chapter 5, we introduce a deep learning approach to achieve film grain rendering. This second contribution boils down to a neural network that approximates the results of a physically based model while greatly accelerating it. We first describe our architecture and its training loss, drawing similarities with previous work on style transfer. In particular, we leverage previous style transfer work which introduced adaptive instance normalisation and make our network dependent on grain size, a physical parameter that highly influences the final aspect of the image. Additionally, this network is designed to be as light as possible, in order to provide faster inference time. We also take care to allow stochastic outputs, which is a key component of film grain. Then, as done in chapter 3, we provide studies and metrics about the performances of our approach, allowing for a comparison with other state-of-the-art methods and showing that our work offers the best tradeoff between quality and speed.Finally, in the last chapter of this thesis, we give a conclusion on our contributions and present potential perspectives following these works.
• Simultaneously transmitting and reflecting reconfigurable intelligent surfaces with energy harvesting from vibrations
  
  • Boujemaa Hatem
  • Alhussein Musaed
  • Rekaya Ghaya
  Signal, Image and Video Processing, Springer Verlag, 2024, 19 (1), pp.84:1-84:8. Simultaneously Transmitting and Reflecting Reconfigurable Intelligent Surfaces (STAR-RIS) represent a cutting-edge technology in wireless communications, allowing surfaces to both transmit and reflect signals simultaneously. This dual functionality provides greater flexibility and efficiency in signal propagation and spectrum management. Integrating energy harvesting from vibrations into STAR-RIS creates an innovative and sustainable solution for powering these systems, enabling autonomous operation in environments where conventional power sources are unavailable. By converting ambient mechanical energy into electrical power, vibration-based energy harvesting supports the continuous operation of STAR-RIS without reliance on external energy sources. This advancement has significant implications for the deployment of future wireless networks, including smart cities, the Internet of Things, and remote sensing applications, offering a pathway to greener and more efficient communication infrastructure. (10.1007/s11760-024-03643-x)
  DOI : 10.1007/s11760-024-03643-x
• Information theory and reinforcement learning of mixed covert and non-covert wireless networks
  
  • Bounhar Abdelaziz
  , 2024. While cryptographic methods offer security, they are often impractical for Internet of Things (IoT) devices due to their limited computational resources and battery life. In light of these challenges, physical layer security techniques, particularly covert communication, seems to be an adequate solution for securing IoT communications. Existing research on covert communication has predominantly focused on systems with solely covert users. This thesis addresses this gap and pioneers the characterization of the information-theoretic fundamental limits of communication systems involving both covert and non-covert users, demonstrating how and when non-covert users can enhance covert communication. It also advances previous findings on the single and multi-users setup by characterizing the exact secret-key rate needed to communicate at a given covert data rate.In another line of work, we address the central approach to modern semantic and goal-oriented communication systems. Specifically, we address the joint source-channel coding problem under a covertness constraints, identifying optimal coding schemes that meet the covertness requirement. These theoretical insights are validated through deep learning techniques, showing that covert semantic communication is only guaranteed when the established theoretical constraints are met. Lastly, to further enrich our research, we extend our work to setups that encompass both covert and non-covert users operating using Non-Orthogonal Multiple Access in an Additive White Gaussian Noise channel. By leveraging reinforcement learning techniques, we develop efficient resource allocation policies that effectively optimize performance in these intricate environments, accounting for real-world constraints such as imperfect channel state information and energy limitations.
• Energy preserving evolutions over Bosonic systems
  
  • Gondolf Paul
  • Möbus Tim
  • Rouzé Cambyse
  Quantum, Verein zur Förderung des Open Access Publizierens in den Quantenwissenschaften, 2024, 8, pp.1551. The exponential convergence to invariant subspaces of quantum Markov semigroups plays a crucial role in quantum information theory. One such example is in bosonic error correction schemes, where dissipation is used to drive states back to the code-space – an invariant subspace protected against certain types of errors. In this paper, we investigate perturbations of quantum dynamical semigroups that operate on continuous variable (CV) systems and admit an invariant subspace. First, we prove a generation theorem for quantum Markov semigroups on CV systems under the physical assumptions that (i) the generator is in GKSL form with corresponding jump operators defined as polynomials of annihilation and creation operators; and (ii) the (possibly unbounded) generator increases all moments in a controlled manner. Additionally, we show that the level sets of operators with bounded first moments are admissible subspaces of the evolution, providing the foundations for a perturbative analysis. Our results also extend to time-dependent semigroups and multi-mode systems. We apply our general framework to two settings of interest in continuous variable quantum information processing. First, we provide a new scheme for deriving continuity bounds on the energy-constrained capacities of Markovian perturbations of quantum dynamical semigroups. Second, we provide quantitative perturbation bounds for the steady state of the quantum Ornstein-Uhlenbeck semigroup and the invariant subspace of the photon dissipation used in bosonic error correction. (10.22331/q-2024-12-04-1551)
  DOI : 10.22331/q-2024-12-04-1551
• Semidefinite programming relaxations for quantum correlations
  
  • Tavakoli Armin
  • Pozas-Kerstjens Alejandro
  • Brown Peter
  • Araújo Mateus
  Reviews of Modern Physics, American Physical Society, 2024, 96 (4), pp.045006. Semidefinite programs are convex optimisation problems involving a linear objective function and a domain of positive semidefinite matrices. Over the last two decades, they have become an indispensable tool in quantum information science. Many otherwise intractable fundamental and applied problems can be successfully approached by means of relaxation to a semidefinite program. Here, we review such methodology in the context of quantum correlations. We discuss how the core idea of semidefinite relaxations can be adapted for a variety of research topics in quantum correlations, including nonlocality, quantum communication, quantum networks, entanglement, and quantum cryptography. (10.1103/RevModPhys.96.045006)
  DOI : 10.1103/RevModPhys.96.045006
• SING: Stability-Incorporated Neighborhood Graph
  
  • Marin Diana
  • Parakkat Amal Dev
  • Ohrhallinger Stefan
  • Wimmer Michael
  • Oudot Steve
  • Memari Pooran
  , 2024, pp.1-10. We introduce the Stability-Incorporated Neighborhood Graph (SING), a novel density-aware structure designed to capture the intrinsic geometric properties of a point set. We improve upon the spheres-of-influence graph by incorporating additional features to offer more flexibility and control in encoding proximity information and capturing local density variations. Through persistence analysis on our proximity graph, we propose a new clustering technique and explore additional variants incorporating extra features for the proximity criterion. Alongside the detailed analysis and comparison to evaluate its performance on various datasets, our experiments demonstrate that the proposed method can effectively extract meaningful clusters from diverse datasets with variations in density and correlation. Our application scenarios underscore the advantages of the proposed graph over classical neighborhood graphs, particularly in terms of parameter tuning. (10.1145/3680528.3687674)
  DOI : 10.1145/3680528.3687674
• Run-Time Adaptation of Neural Beamforming for Robust Speech Dereverberation and Denoising
  
  • Fujita Yoto
  • Nugraha Aditya Arie
  • Carlo Diego Di
  • Bando Yoshiaki
  • Fontaine Mathieu
  • Yoshii Kazuyoshi
  , 2024. This paper describes speech enhancement for realtime automatic speech recognition (ASR) in real environments. A standard approach to this task is to use neural beamforming that can work efficiently in an online manner. It estimates the masks of clean dry speech from a noisy echoic mixture spectrogram with a deep neural network (DNN) and then computes a enhancement filter used for beamforming. The performance of such a supervised approach, however, is drastically degraded under mismatched conditions. This calls for run-time adaptation of the DNN. Although the ground-truth speech spectrogram required for adaptation is not available at run time, blind dereverberation and separation methods such as weighted prediction error (WPE) and fast multichannel nonnegative matrix factorization (FastMNMF) can be used for generating pseudo groundtruth data from a mixture. Based on this idea, a prior work proposed a dual-process system based on a cascade of WPE and minimum variance distortionless response (MVDR) beamforming asynchronously fine-tuned by block-online FastMNMF. To integrate the dereverberation capability into neural beamforming and make it fine-tunable at run time, we propose to use weighted power minimization distortionless response (WPD) beamforming, a unified version of WPE and minimum power distortionless response (MPDR), whose joint dereverberation and denoising filter is estimated using a DNN. We evaluated the impact of run-time adaptation under various conditions with different numbers of speakers, reverberation times, and signal-to-noise ratios (SNRs).
• Dynamic Analysis of Influencer Impact on Opinion Formation in Social Networks
  
  • Berjawi Omran
  • Cavaliere Danilo
  • Fenza Giuseppe
  • Khatoun Rida
  , 2025, 15463, pp.394–408. The rapid proliferation of social media platforms has transformed communication, enabling individuals to share opinions and influence others on an unprecedented scale. This paper addresses the challenge of quantifying the ability of social media influencers to change opinions over time. Traditional metrics, such as follower counts or engagement rates, offer a limited view of an influencer’s true impact. To face this challenge, this study provides a nuanced framework based on Friedkin-Johnsen model and Sentiment Analysis for analyzing how people’s opinions propagate through social networks and how influencers can affect these dynamics. The methodology consists in building interaction network graphs, detecting communities, and identifying key influencers using classic topology metrics. Then, it applies Sentiment Analysis to capture users’ opinions, which are injected into the Friedkin-Johnsen model to study their evolution over time. The results show the effectiveness of the proposed approach in determining the dynamics of social influence and opinion change. (10.1007/978-981-96-1483-7_32)
  DOI : 10.1007/978-981-96-1483-7_32
• Statistical Learning for Spatial data : theory and practice
  
  • Siviero Emilia
  , 2024. In the Big Data era, massive datasets exhibiting a possibly complex spatial dependence structure are becoming increasingly available. In this thesis, we aim at developing approaches to efficiently exploit the dependence structure of spatial (and spatio-temporal) data.We first analyze the simple Kriging task, the flagship problem in Geostatistics, from a statistical learning perspective, i.e. by carrying out a non-parametric finite-sample predictive analysis. In this context, the standard probabilistic theory of statistical learning does not apply directly and theoretical guarantees of the generalization capacity of the Kriging predictive rule learned from spatial data are left to be established. Given a finite number of values taken by a realization of a square integrable random field, with unknown covariance structure, the goal is to predict the unknown values that the random field takes at any other location in the spatial domain with minimum quadratic risk. Establishing the generalization capacity of empirical risk minimizer is far from straightforward, due to the non independent and identically distributed nature of the training data involved in the learning procedure. In the first part of this thesis, non-asymptotic bounds 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 in the learning stage. These theoretical results, as well as the role played by the technical conditions required to establish them, are illustrated by various numerical experiments, on simulated data and on real-world datasets, and may hopefully pave the way for further developments in statistical learning based on spatial data.In the second part of this thesis, we focus on space-time Hawkes processes. Many modern spatio-temporal data sets, in sociology, epidemiology or seismology, for example, exhibit self-exciting characteristics, with simultaneous triggering and clustering behaviors, that a suitable spatio-temporal Hawkes process can accurately capture.However, dealing efficiently with the high volumes of data now available is challenging.We aim at developing a fast and flexible parametric inference technique to recover the parameters of the kernel functions involved in the intensity function of a spatio-temporal Hawkes process based on such data. Our statistical approach combines three key ingredients: (1) kernels with finite support are considered, (2) the space-time domain is appropriately discretized, and (3) (approximate) precomputations are used. The inference technique we propose consists of a fast and statistically accurate solver.In addition to describing the algorithmic aspects, numerical experiments have been carried out on synthetic and real spatio-temporal data, providing solid empirical evidence of the relevance of the proposed methodology.
• ADEPT 2023 Workshop Summary
  
  • Tran Hai Nam
  • Singhoff Frank
  • Hugues Jérôme
  • Dissaux Pierre
  • Lewis Bruce
  • Shackleton Hazel
  • Kiniry Joseph
  • Zeyda Frank
  • Rakshit Mittal
  • Blouin Dominique
  • Bhobe Anish
  • Pautet Laurent
  • Bae Kyungmin
  • Ölveczky Peter Csaba
  • Larson Brian R
  • Ahmadifar Ehsan
  • Kosmidis Leonidas
  • Valente Hugo
  • a Miguel Miguel
  • G Perez Ángel
  • Perez Alonso Alejandro
  • Zamorano Juan
  • Antonio de La Puente Juan
  Ada Letters, Association for Computing Machinery, 2024, 44 (1), pp.23-25. The Architecture Analysis and Design Language (AADL) is a SAE standard for modeling both hardware and software architecture of embedded systems. Widely embraced by stakeholders in critical real-time embedded systems, the AADL standard is used to address a large set of concerns including performances (latency, schedulability), safety, and security. The ADEPT workshop aims to present and report on current projects in the field of design, implementation, and verification of critical real-time embedded systems where AADL is a first-citizen technology. This article is a summary of the second edition of the workshop in 2023. (10.1145/3706601.3706603)
  DOI : 10.1145/3706601.3706603
• OVOSE: Open-Vocabulary Semantic Segmentation in Event-Based Cameras
  
  • Rahman Muhammad Rameez Ur
  • Giraldo Jhony H.
  • Spinelli Indro
  • Lathuilière Stéphane
  • Galasso Fabio
  , 2024, 15316, pp.18–33. Event cameras, known for low-latency operation and superior performance in challenging lighting conditions, are suitable for sensitive computer vision tasks such as semantic segmentation in autonomous driving. However, challenges arise due to limited event-based data and the absence of large-scale segmentation benchmarks. Current works are confined to closed-set semantic segmentation, limiting their adaptability to other applications. In this paper, we introduce OVOSE, the first Open-Vocabulary Semantic Segmentation algorithm for Event cameras. OVOSE leverages synthetic event data and knowledge distillation from a pre-trained image-based foundation model to an event-based counterpart, effectively preserving spatial context and transferring open-vocabulary semantic segmentation capabilities. We evaluate the performance of OVOSE on two driving semantic segmentation datasets DDD17, and DSEC-Semantic, comparing it with existing conventional image open-vocabulary models adapted for event-based data. Similarly, we compare OVOSE with state-of-the-art methods designed for closed-set settings in unsupervised domain adaptation for event-based semantic segmentation. OVOSE demonstrates superior performance, showcasing its potential for real-world applications. The code is available at https://github.com/ram95d/OVOSE. (10.1007/978-3-031-78444-6_2)
  DOI : 10.1007/978-3-031-78444-6_2
• WaterMAS: sharpness-aware maximization for neural network watermarking
  
  • de Sousa Trias Carl
  • Mitrea Mihai
  • Fiandrotti Attilio
  • Cagnazzo Marco
  • Chaudhuri Sumanta
  • Tartaglione Enzo
  , 2024, 15305, pp.301-317. Nowadays, deep neural networks are used for solving complex tasks in several critical applications and protecting both their integrity and intellectual property rights (IPR) has become of utmost importance. To this end, we advance WaterMAS, a substitutive, white-box neural network watermarking method that improves the trade-off among robustness, imperceptibility, and computational complexity, while making provisions for increased data payload and security. WasterMAS insertion keeps unchanged the watermarked weights while sharpening their underlying gradient space. The robustness is thus ensured by limiting the attack’s strength: even small alterations of the watermarked weights would impact the model’s performance. The imperceptibility is ensured by inserting the watermark during the training process. The relationship among the WaterMAS data payload, imperceptibility, and robustness properties is discussed. The secret key is represented by the positions of the weights conveying the watermark, randomly chosen through multiple layers of the model. The security is evaluated by investigating the case in which an attacker would intercept the key. The experimental validations consider 5 models and 2 tasks (VGG16, ResNet18, MobileNetV3, SwinT for CIFAR10 image classification, and DeepLabV3 for Cityscapes image segmentation) as well as 4 types of attacks (Gaussian noise addition, pruning, fine-tuning, and quantization). The code will be released open-source upon acceptance of the article. (10.1007/978-3-031-78169-8_20)
  DOI : 10.1007/978-3-031-78169-8_20