Publications

2026

• Real-World Assessment of RF-EMF Exposure From 3G, 4G, and 5G Mobile Networks Across Different Urbanization Levels
  
  • Lee Ae-Kyoung
  • Jeon Sangbong
  • Hong Seon-Eui
  • Wang Shanshan
  • Wiart Joe
  • Samaras Theodoros
  • Moon Jung Ick
  IEEE Access, IEEE, 2026, 14, pp.134636 - 134653. As 5G mobile networks become increasingly widespread, realistic assessment of radio-frequency electromagnetic field (RF-EMF) exposure in modern mobile-network environments is becoming increasingly important. This study characterized real-world downlink (DL) and uplink (UL) RF-EMF exposure across different urbanization levels in South Korea using commercial 3G, 4G, and 5G mobile networks. DL electric-field strengths and UL transmit powers were simultaneously measured in urban and rural environments using frequency-selective field measurements and mobile-network logging systems. The measured exposure levels were used to calculate brain specific absorption rate (SAR) using anatomically realistic child and adult numerical models. The results showed that, under the measurement and dosimetric conditions considered in this study, brain SAR associated with DL exposure from base-station infrastructure was comparable to or exceeded that associated with UL exposure in most tested scenarios. Urban–rural differences in whole-brain-averaged SAR reached 6.2 dB for the child model and 5.7 dB for the adult model, whereas age-related differences were substantially smaller. These findings indicate that environmental and network-infrastructure factors are major determinants of RF-EMF exposure variability in dense modern mobile-network environments. By integrating real-world measurements with anatomically realistic dosimetry, this study provides a practical framework for RF-EMF exposure assessment and characterization in evolving mobile communication networks. (10.1109/ACCESS.2026.3728447)
  DOI : 10.1109/ACCESS.2026.3728447
• RAVEN: A Virtual Platform for RISC-V Microarchitectural Security
  
  • Khan Mahreen
  • Mushtaq Maria
  • Pacalet Renaud
  • Apvrille Ludovic
  ACM Transactions on Modeling and Computer Simulation, Association for Computing Machinery, 2026, 36 (4), pp.1-27. <div><p>The rise of the open-source RISC-V Instruction Set Architecture (ISA) has highlighted the critical need for a standardized framework for microarchitectural security evaluation, a gap that challenges the development of secure-by-design hardware. To address this, we present a novel virtual simulation platform, RAVEN, for evaluating the microarchitectural security of RISC-V processors. Built on the gem5 simulator, the proposed framework enables detailed modeling of cache hierarchies, integration of custom hardware performance counters, and fine-grained analysis of timing-based side-channel vulnerabilities. The platform is designed to support both attack emulation and automated detection, providing a flexible environment for architecture-level security evaluation. We validate the platform by implementing and analyzing five representative microarchitectural attacks: Spectre, Evict+Time, Evict+Spec+Time, Flush+Fault, and GhostRace. To detect these attacks, we integrate a machine learning-based detection pipeline that leverages hardware performance counter data to identify anomalous microarchitectural behavior. Our experimental results across multiple classifiers show high detection accuracy, demonstrating the platform's effectiveness in distinguishing side-channel activity from benign execution.</p><p>Overall, the extensible gem5-based platform provides a practical and scalable foundation for security research on the open RISC-V ISA, supporting both attack analysis and the development of secure-by-design hardware.</p></div> <div>CCS Concepts: •</div> (10.1145/3811543)
  DOI : 10.1145/3811543
• On the privacy cost for dependent Gaussian data: spectral density estimation under local differential privacy
  
  • Issartel Yann
  • Roueff François
  , 2026. We study the fundamental problem of estimating the dependence structure of a centered stationary Gaussian process under local differential privacy (LDP). In this setting, the spectral density characterizes the dependence structure of the data and is the quantity to be estimated. Our main contribution is to close the open $\alpha^2$-versus-$\alpha^4$ gap between the previously known lower and upper bounds on the minimax rate. Specifically, we establish a minimax lower bound showing that, over Sobolev-type classes of spectral densities, the effective sample size in the high-privacy regime is $N\alpha^4$, rather than the usual $N\alpha^2$ arising for independent observations. This additional privacy cost is caused by the temporal dependence between the observations rather than by their marginal distributions. The proof relies on a contraction bound for privatized dependent Gaussian observations. Our second contribution is a matching upper bound, free of the polylogarithmic losses present in previous work. Rather than applying a generic privatization scheme to classical estimators, we construct a problem-specific procedure attaining the rate identified by our lower bound. Beyond closing the gaps in spectral density estimation, we apply the tools developed for this problem to several related questions. We (i) close the logarithmic gap for fixed-lag autocovariance estimation, (ii) show that the $\alpha^4$ cost arises locally around every spectral density bounded away from zero, and (iii) establish that classical asymptotic equivalence with an independent Gaussian experiment generally fails under LDP.
• RISC‑V Microarchitectural Attacks: No Timer, No Problem?
  
  • Khan Mahreen
  • Bois--Baumann Gaetan
  • Mushtaq Maria
  • Pacalet Renaud
  • Apvrille Ludovic
  IEEE Embedded Systems Letters, Institute of Electrical and Electronics Engineers, 2026, pp.1-1. RISC-V processors are increasingly being deployed in cyber-physical, time-sensitive embedded, and reactive systems, making microarchitectural timing side-channel attacks a practical security concern. A commonly proposed countermeasure is to disable access to the high-resolution rdcycle timer under the assumption that practical attacks require precise timing measurements. However, this approach is unsuitable for systems that rely on precise timing for scheduling and deadline enforcement, and it does not remove the underlying secret-dependent microarchitectural state. We demonstrate the limitations of this countermeasure by evaluating four cache and transient-execution attacks, namely Flush+Reload, Flush+Flush, Flush+Fault, and Spectre V1, on a commercial out-of-order RISC-V core, the T-Head C910. We replace rdcycle with three alternative timing sources: a software counter thread, clock_gettime, and perf_event_open. Under background CPU loads ranging from 25% to 100%, with 50 repetitions per configuration, the alternative timing sources achieve up to a 93% average attack success rate across the evaluated settings, showing that the attacks remain effective even under substantial system noise. We further demonstrate complete 128-bit AES key recovery, clear timing separation between page-table-walk and no-walk executions, and a cross-core covert channel that transmits 1,024 bits with zero errors in 49 out of 50 independent runs. We also evaluate index masking and speculation-fence insertion for Spectre V1 and examine L1 instruction-cache and L1 data-cache miss rates as indicators for profiling-based attack detection. These results show that restricting rdcycle alone is insufficient and that effective protection requires layered hardware and software countermeasures. (10.1109/LES.2026.3726619)
  DOI : 10.1109/LES.2026.3726619
• Exploiting Subgradient Sparsity in Max-Plus Neural Networks
  
  • Enaieh Ikhlas
  • Fercoq Olivier
  , 2026. Deep Neural Networks are powerful tools for solving machine learning problems, but their training often involves dense and costly parameter updates. In this work, we use a novel Max-Plus neural architecture in which classical addition and multiplication are replaced with maximum and summation operations respectively. This is a promising architecture in terms of interpretability, but its training is challenging. A particular feature is that this algebraic structure naturally induces sparsity in the subgradients, as only neurons that contribute to the maximum affect the loss. However, standard backpropagation fails to exploit this sparsity, leading to unnecessary computations. In this work, we focus on the minimization of the worst sample loss which transfers this sparsity to the optimization loss. To address this, we propose a sparse subgradient algorithm that explicitly exploits the algebraic sparsity. By tailoring the optimization procedure to the non-smooth nature of Max-Plus models, our method achieves more efficient updates while retaining theoretical guarantees. This highlights a principled path toward bridging algebraic structure and scalable learning.
• Neural Multichannel Distant Speaker Diarization and Source Separation with Beta Speaker Activity Prior
  
  • Mao Sicheng
  • Fontaine Mathieu
  • Larcher Anthony
  • Badeau Roland
  , 2026. Distant speaker diarization remains challenging due to adverse acoustic conditions, varying numbers of speakers and overlapping speech. While data-driven approaches have shown strong performance, model-driven methods offer a compelling alternative by leveraging spatial information from multichannel recordings. This paper is motivated to propose a Bayesian diarization model for a model-driven method called neural FCASA to enhance its robustness. Specifically, we propose a beta prior over speaker activity and hence a variational lower bound objective that can be seen as a regularized continuous speaker activity score in place of the original cross-entropy loss to train the diarization model. Our experiments show significant improvements in terms of Diarization Error Rate by at least 3% (16% relatively) and Jaccard Error Rate by at least 4% (20% relatively) on the AMI dataset compared to the baseline.
• Technical Report NF-PERSEUS 2024
  
  • Abdel Nour Charbel
  • Adjih Cedric
  • Hamzah Amazig Anis
  • Amis Karine
  • Bégaud Xavier
  • Berro Rim
  • Cances Jean-Pierre
  • Collardey Sylvain
  • Crussière Matthieu
  • Dabouz Mounir
  • Delaveaud Christophe
  • Demmer David
  • Doumit Joseph
  • Di Renzo Marco
  • Douillard Catherine
  • Durant Antoine
  • El Hassani Hajar
  • Farah Joumana
  • Gaillot Davy
  • Giry Alexandre
  • Goursaud Claire
  • Guillaud Maxime
  • Hamdar Fatima
  • Yassine Hentati Mohamed
  • Jafaar Raneem
  • Lepage Anne-Claire
  • Le Ruyet Didier
  • Mabrouk Asma
  • Mannoni Valérian
  • Meghdadi Vahid
  • Nadal Jérémy
  • Nguyen Nhu-Huan
  • Pagani Pascal
  • Dang-Kien Pham Germain
  • Piron Romain
  • Prabhu Balakrishna
  • Sharaiha Ala
  • Simon Eric Pierre
  • Xavier Pascal
  , 2026. This is the second deliverable D2 "Technical Report NF-PERSEUS 2024". This document provides an overview of the progress of the studies undertaken in NF-PERSEUS project and their progress by the end of 2024.</p><p>Chapter 1 delves into the advantages of cell-free architectures and their pivotal role in shaping the future of 6G wireless networks. Cell-free architectures, which move away from the traditional cellular network model, offer several compelling benefits that make them a promising solution for the next generation of mobile communication.</p><p>Chapter 2 describes the main use cases identified for the NF-PERSEUS project and their KPI, providing an inventory of use case specifications, representative deployment scenarios and technical requirements.</p><p>Chapter 3 examines and discusses relevant reference scenarios based on RIS. This analysis sheds light on the technical challenges involved and the potential performance that can be achieved.</p><p>Chapter 4 presents three physical layer innovations for enhancing cell-free massive MIMO systems. It introduces hardware-aware distributed precoding to mitigate nonlinear distortion, optimized filtering for QAM-FBMC to reduce self-interference, and OS-based FBMC receivers that improve resilience to timing and frequency offsets. Together, these contributions aim to increase spectral efficiency and robustness in next-generation wireless networks.</p><p>Chapter 5 presents recent advances in resource management for cell-free massive MIMO systems, developed within WP4 of the NF-PERSEUS project. It covers novel algorithms for resource allocation and non-orthogonal multiple access, as well as grant-free access schemes leveraging AI/ML techniques. A global approach to orchestration and multi-antenna processing is also proposed to optimize performance and energy efficiency in distributed network architectures.</p><p>Chapter 6 addresses the need for revisiting propagation models and radio channel assumptions in the context of upcoming 5G technologies, specifically in dynamic environments like vehicular communications. The work focuses on understanding the non-stationarity of MIMO channels, particularly in cell-free networks, by investigating the correlation matrix distance (CMD) metric. Future research will analyze extensive data from dual-band campaigns to explore time, delay, and angular characteristics in V2I scenarios, aiming to improve adaptive transmission schemes and beamforming algorithms for robust performance in non-stationary environments.</p><p>Chapter 7 explores the emerging role of Reconfigurable Intelligent Surfaces (RIS) in wireless communications, focusing on key challenges and models. It begins with an overview of electromagnetically consistent communication models for RIS, followed by a discussion on multiport network theory for system optimization. The chapter highlights the importance of channel state information in RIS optimization, comparing short-term and long-term design criteria, and concludes by examining the optimization of effective rank and NOMA-aided RIS systems. Through these contributions, the chapter aims to improve the efficiency and performance of RIS-assisted wireless networks.</p><p>Chapter 8 presents the advancements made in the development of innovative antenna technologies within the PERSEUS project, focusing on radio-frequency front-end modules and reconfigurable antennas for massive MIMO (mMIMO) systems. The research explores several approaches, including the integration of reconfigurable planar metasurfaces with varicap diodes for dynamically steerable beams, and compact superdirective antennas embedded in traditional array architectures to enhance beamforming and reduce physical footprint. Additionally, the chapter investigates the use of bio-sourced materials in antenna design, aiming to offer environmentally sustainable solutions without compromising electromagnetic performance. These innovative antenna systems will be evaluated through a demonstrator, allowing real-world testing of the proposed technologies in a cell-free massive MIMO context, contributing to the development of next-generation wireless networks.</p><p>Chapter 9 gives the conclusion.
• Semi-Supervised Novel Intrusion Detection via Orthogonal Feature Extraction and Score-Space Meta-Classifier Assembly
  
  • Vu Minh-Khanh
  • Ene Cristian
  • Mounier Laurent
  • Ramparison Mathias
  , 2026. Signature-based endpoint detection fails against novel payloads whose behavioral patterns do not match any catalogued specimen. A semi-supervised host-based intrusion detection system is introduced that decomposes each system-call window into three structurally independent views, each trained exclusively on benign traces: frequency and rarity (TF-IDF with a one-class Support Vector Machine), local topology (a 1D convolutional autoencoder), and chronological grammar (a GRU next-event predictor). The three anomaly scalars are assembled into a score-space meta-matrix over which an XGBoost layer learns the final decision boundary without accessing raw syscall sequences or malware signatures directly. On the ADFA-LD 32-bit benchmark, the fusion model achieves Area Under the ROC Curve (AUC) 0.9113, Area Under the Precision-Recall Curve (AUPR) 0.8950, and F1 0.8523 at a full-pipeline latency of 0.2217 ms per window. Per-family evaluation across six attack families yields average AUC 0.9073, and cross-path Pearson correlations of 0.43 to 0.46 indicate that the three views occupy productively distinct operating regimes and increase the number of behavioural constraints an attacker must preserve to evade detection.
• Technical Report NF-PERSEUS 2025
  
  • Charbel Abdel
  • Demmer David
  • Renzo Marco Di
  • Douillard Catherine
  • Farah Joumana
  • Gaillot Davy
  • Giry Alexandre
  • Goursaud Claire
  • Guzzo Marco
  • Lepage Anne-Claire
  • Nguyen Nhu Huan
  • Adjih Cédric
  • Amis Karine
  • Ariaudo Myriam
  • Begaud Xavier
  • Berro Rim
  • Cances Jean-Pierre
  • Collardey Sylvain
  • Crussière Matthieu
  • Delavaud Christophe
  • Desombre Paul
  • Durant Antoine
  • Guillaud Maxime
  • Hentati Mohamed Yassine
  • Jaafar Raneem
  • Le Ruyet Didier
  • Asma Mabrouk
  • Mannoni Valerian
  • Medrar Ghiles
  • Meghdadi Vahid
  • Germain Pham Dang-Kièn
  • Poulliat Charly
  • Prabhu Balakrishna
  • Rekaya Ghaya
  • Requena Florian
  • Simon Eric Pierre
  • Vuong Tan-Phu
  • Xavier Pascal
  , 2026. This deliverable D3 "Technical Report NF-PERSEUS 2025" provides an overview of the scientific and technological progress achieved within the NF-PERSEUS project during 2025. It covers the main research activities conducted across the project, from use cases and channel modeling to advanced physical-layer solutions, radio resource management, RIS technologies, and RF front-end and antenna developments.</p><p>Chapter 1 introduces the principles and potential of cell-free massive MIMO architectures for future wireless networks. It discusses their main advantages and the challenges associated with distributed deployments, providing the general framework for the research activities conducted within NF-PERSEUS.</p><p>Chapter 2 presents the main use cases considered within NF-PERSEUS, including ultra-dense areas with very high data rates, smart factories and Vehicle-to-Infrastructure (V2I) communications. The chapter describes the corresponding deployment scenarios with technical requirements and highlights the potential of cell-free massive MIMO to provide homogeneous coverage, high capacity, scalability and reliable communications.</p><p>Chapter 3 focuses on RIS-enabled scenarios and use cases. Based on the recommendations and activities of ETSI ISG RIS, it identifies relevant deployment configurations and discusses the potential benefits of RIS in terms of coverage, spectral efficiency, beam management, interference mitigation and multiple access. These scenarios provide the basis for the RIS-related research conducted within the project.</p><p>Chapter 4 presents the main advances achieved in NF-PERSEUS on advanced PHY-layer solutions. The contributions cover hardware-aware end-to-end learning, PAPR analysis and control, low-complexity signal processing, FBMC waveforms and pilot design, channel estimation, channel coding and multi-user detection. The results demonstrate the potential of these approaches to improve spectral and energy efficiency, robustness and implementation complexity in distributed massive MIMO systems.</p><p>Chapter 5 reports the research activities on radio resource management. It investigates advanced NOMA schemes, predictive resource allocation, AI/ML-based access and resource management, quantum-inspired detection, RIS-assisted transmission and joint optimization of user association, precoding and interference management. These contributions address the scalability, fairness and energy-efficiency challenges associated with massive access in distributed B5G/6G networks.</p><p>Chapter 6 presents considerations for massive MIMO channel modeling and reports results from channel measurement campaigns conducted for cell-free V2I scenarios. In particular, the measurements provide experimental data on polarization, spatial characteristics and propagation mechanisms in distributed deployments, supporting the development of realistic channel models for the evaluation of NF-PERSEUS technologies.</p><p>Chapter 7 investigates modeling, performance evaluation and optimization of Reconfigurable Intelligent Surfaces. It reviews electromagnetically consistent RIS communication models and develops optimization approaches based on multiport network representations. The chapter also studies the impact of channel state information and channel estimation errors on RIS optimization as well as RIS-assisted multi-user and NOMA systems.</p><p>Chapter 8 summarizes preliminary results on RF front-end modules, reconfigurable antennas and RIS technologies. The reported developments include energy-efficient power amplifiers, digital pre-distortion, compact and reconfigurable antenna architectures, Huygens metasurfaces, beam-switchable Yagi-Uda arrays and bio-sourced antenna materials. These developments target the reduction of power consumption, improved adaptability and compact integration of radio technologies for future distributed massive MIMO systems.</p><p>Chapter 9 positions representative NF-PERSEUS results with respect to the state of the art. The benchmarking highlights quantified improvements in performance, complexity, robustness and energy efficiency across RF and antenna technologies, PHY-layer solutions and radio resource management. It also demonstrates the increasing consideration of practical constraints, including hardware impairments, imperfect channel knowledge, dense deployments and computational complexity.</p><p>Overall, the results presented in this deliverable demonstrate significant progress towards the development of scalable, energy-efficient and sustainable wireless technologies for future 6G networks. The complementary contributions developed across the different layers of the system, from RF front-ends and programmable propagation environments to PHY-layer processing and radio resource management, provide a coherent technological foundation for further integration, experimental validation and maturation towards higher TRLs.
• Superviz26-SQL: A Multi-Domain Benchmark for SQL Attack Detection
  
  • Quetel Grégor
  • Gimenez Pierre-François
  • Robert Thomas
  • Pautet Laurent
  , 2026. Machine-learning intrusion detectors that excel on a testbed often fail once the deployment environment changes. Yet comparable multi-domain datasets with controlled distribution shifts remain scarce. This holds for SQL attack detection (SQLAD), whose public clear-text data covers only single applications, leaving the cross-domain behaviour of detection methods unmeasured. We introduce Superviz26-SQL, the first multi-domain SQLAD benchmark, extending the synthetic generation methodology of Superviz25-SQL to four heterogeneous domains derived from real-world database projects. From these domains we derive ready-to-use datasets for three evaluation protocols: Superviz26-SQL-LODO for cross-domain generalisation, Superviz26-SQL-CD for samedomain concept drift, and Superviz26-SQL-FSL for few-shot adaptation. Lexical-field analysis confirms genuine inter-domain diversity, and each training set is large enough to train deep detectors. We illustrate each protocol using reference baselines spanning five feature extractors and three decision engines. Superviz26-SQL, its three derivatives, and all generation and experiment code are publicly released to support reproducible cross-domain SQLAD research.
• How Predicted Links Influence Network Evolution: Disentangling Choice and Algorithmic Feedback in Dynamic Graphs
  
  • Perez Mathilde
  • Romero Raphaël
  • Lijffijt Jefrey
  • Laclau Charlotte
  , 2026. Link prediction models are increasingly used to recommend interactions in evolving networks, yet their impact on network structure is typically assessed from static snapshots. In particular, observed homophily conflates intrinsic interaction tendencies with amplification effects induced by network dynamics and algorithmic feedback. We propose a temporal framework based on multivariate Hawkes processes that disentangles these two sources and introduce an instantaneous bias measure derived from interaction intensities, capturing current reinforcement dynamics beyond cumulative metrics. We provide a theoretical characterization of the stability and convergence of the induced dynamics, and experiments show that the proposed measure reliably reflects algorithmic feedback effects across different link prediction strategies.
• CutClean: Neural Network Pruning for Privacy-Preserving Inference
  
  • Magliolo Leonardo
  • Pastore Vito Paolo
  • Valenzise Giuseppe
  • Tartaglione Enzo
  , 2026. <div><p>Neural networks are increasingly deployed in high-stakes applications with growing privacy leakage concerns. We show that this privacy leakage can occur even in the absence of representation imbalances that lead to traditional dataset biases. This poses significant privacy risks when deploying models that process sensitive attributes. In this context, we propose CutClean, a privacy-aware pruning method that allows to reduce privacy information flow through the network, while increasing its sparsity. Our approach employs auxiliary linear privacy heads placed at each network's block to quantify information leakage, and further applies increasing levels of sparsity to remove the private attribute leakage, measured in terms of the accuracy of the privacy head attached to the last block. Experiments on synthetic and real-world datasets demonstrate that our approach effectively minimizes private information flow while achieving high sparsity rates and preserving classification target accuracy.</p></div>
• Learning and simulating bosonic systems via finite-energy locality
  
  • Möbus Tim
  • Bluhm Andreas
  • Caro Matthias
  • Werner Albert
  • Rouzé Cambyse
  , 2026. Bosonic devices promise applications in simulation, sensing and quantum error correction, but infinite-dimensional local Hilbert spaces obstruct the locality tools that make qubit dynamics efficiently learnable and simulable. We establish a finite-energy locality principle for geometrically local bosonic open systems satisfying photon-number moment propagation. It compares unbounded generators with Galerkin cutoffs, transferring finite-dimensional Lieb--Robinson, product-formula and circuit techniques to bosons with explicit errors. For this moment-controlled class, we obtain, to our knowledge, the first model-independent weak Lieb--Robinson bounds beyond Bose--Hubbard-type dynamics, together with quantitative Trotter and simulation guarantees for polynomial bosonic GKSL generators. As a central application, coherent-state preparation and local heterodyne detection suffice to learn coefficients of a known bounded-degree polynomial Hamiltonian ansatz local on a bounded-growth interaction graph to accuracy $\varepsilon$ and failure probability $\delta$, with sample complexity and total evolution time both $\widetilde{\mathcal{O}}(\varepsilon^{-2}\log(m/\delta))$, where $m$ counts on-site and interaction terms. This matches the best known finite-dimensional locality-assisted scaling in accuracy and system size, up to polylogarithmic factors. The assumptions hold for Bose--Hubbard and quadratic dynamics without added dissipation; for more general local polynomial Hamiltonians they can be supplied natively or engineered by known multi-photon loss in stabilized bosonic architectures. (10.48550/arXiv.2307.15026)
  DOI : 10.48550/arXiv.2307.15026
• ReLTEx: Reliable LLM-based Taxonomy Expansion
  
  • Ghamlouch Zeinab
  • Alam Mehwish
  , 2026. Recent advances in Large Language Models (LLMs) have demonstrated strong capabilities in generating semantically relevant concepts and relations, making them promising tools for taxonomy enrichment. However, directly relying on LLM-generated expansions often leads to noisy, redundant, or hierarchically inconsistent structures, limiting their reliability for automated taxonomy expansion. In this paper, we present ReLTEx, a framework for reliable LLM-based taxonomy expansion. ReLTEx combines LLM-driven candidate generation with structure-aware validation and recursive expansion control to improve the consistency and quality of generated taxonomies by reducing hallucinations. We evaluate the proposed framework using benchmark taxonomies under a masked taxonomy expansion setting and compare multiple validation strategies. Experimental results, supported by both adapted evaluation metrics and human evaluation, demonstrate that ReLTEx produces more reliable and semantically coherent taxonomy expansions.
• LELA: An End-to-end LLM-based Entity Linking Framework with Zero-shot Domain Adaptation
  
  • Haffoudhi Samy
  • Dobričić Nikola
  • Suchanek Fabian
  • Holzenberger Nils
  , 2026. Entity linking is a key component of many downstream NLP systems, yet existing approaches are often tied to the specific target knowledge bases and domains, limiting their real world application. In this paper, we extend LELA, a modular and domain-agnostic LLM-based entity disambiguation method, into a practical Python library that integrates zero-shot Named Entity Recognition (NER) -thereby providing a complete end-toend pipeline for entity-linking in real-world usage. We provide experimental results validating LELA's performance and robustness across diverse entity linking settings. In our demo, users can play with the system on their own input texts.
• Evaluation of Electromagnetic Exposure to Signal-to-Interference Ratio in Downlink Cellular Networks
  
  • Nguyen Minh-Huy
  • Wang Shanshan
  • Wiart Joe
  , 2026, pp.1-4. This paper presents a stochastic geometry based framework for analyzing downlink (DL) Electromagnetic Field (EMF) exposure and the Signal-to-Interference Ratio (SIR) in urban environments. In communication systems, the EMFexposure measures the absorption of electric and magnetic field energy emanating from communication devices, while SIR reflects link reliability and communication quality. These two metrics are usually analyzed distinctly, leaving the evaluation of the trade-off between these two insufficiently addressed. A parameter E-SIR (exposure-to-SIR) is introduced to demonstrate the ratio between exposure and signal quality. The E-SIR cumulative distribution function (CDF) is derived for a general case, and a special case with path loss exponent α = 4, validated by numerical results. The result shows an agreement in the numerical simulation and the analytical framework. This enables the evaluation of EMF exposure and signal quality trade-off, quantifying the probability that a user achieves a target SIR while keeping EMF exposure below a safety limit under the network densification scenario. The decrement of the E-SIR CDFwhen base stations (BSs) becomedenser indicates that users are more likely to experience higher exposure relative to the achieved SIR. (10.46620/URSIGASS26/WXQH4454)
  DOI : 10.46620/URSIGASS26/WXQH4454
• Characterizing Perceived Readability in Data Visualization: Design and Reader Factors
  
  • Cabouat Anne-Flore
  • Kurzhals Kuno
  • Matzen Laura
  • Isenberg Tobias
  • Huron Samuel
  • Isenberg Petra
  , 2026. We report results from an online study designed to characterize perceived readability across a controlled set of static visual data representations.
• On the hydrodynamic behaviour of the immersed boundary-lattice Boltzmann method for wetting problems (a)
  
  • Bellantoni Elisa
  • Guglietta Fabio
  • Demou Andreas
  • Pelusi Francesca
  • Um Kiwon
  • Nicolaou Mihalis
  • Desbrun Mathieu
  • Sbragaglia Mauro
  • Savva Nikos
  EPL - Europhysics Letters, European Physical Society / EDP Sciences / Società Italiana di Fisica / IOP Publishing, 2026, 155 (6), pp.63002. We study the hydrodynamic behaviour of a mesoscale numerical model for wetting dynamics based on the immersed boundary-lattice Boltzmann (IBLB) method. This IBLB model features a wetting potential to capture the interaction between a non-ideal droplet interface and a solid boundary; it is designed to prevent abrupt curvature changes near the contact line. As this approach prevents direct contact between the droplet and the solid, it forms a thin film beneath the droplet, which could compromise the hydrodynamic consistency in this region. This paper presents detailed comparisons against two other hydrodynamic solvers, respectively based on a boundary element method (BEM) and a volume of fluid (VoF) method, in order to examine the hydrodynamic behaviour of this IBLB scheme, elucidate its limits of validity in wetting applications, and explore the properties of its contact-line model. (10.1209/0295-5075/ae92af)
  DOI : 10.1209/0295-5075/ae92af
• Self-supervised interferogram restoration by regularized inversion of despeckled projections
  
  • Gaya Victor
  • Denis Loïc
  • Pinel-Puysségur Béatrice
  • Bultingaire Thomas
  • Tupin Florence
  IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing, IEEE, 2026, 19, pp.28310 - 28324. Synthetic Aperture Radar Interferometry (InSAR) gives access to topographical and displacement information. Yet, the estimation of interferometric phase and coherence requires refined image restoration techniques to reduce fluctuations due to spatial and temporal decorrelations. Recently, the generic approach MuChaPro has been proposed to reduce speckle in multi-channel SAR data by reducing, thanks to projections, the restoration problem to a series of single-channel despeckling tasks. In the specific case of interferometric data, this method shows some limits in very low coherence areas. This paper identifies these limitations and proposes to improve MuChaPro estimation by introducing a spatial regularization framework dedicated to interferometric data. Experiments on simulated interferograms and real TerraSAR-X and PAZ data show that the proposed optimization framework significantly improves the quality of the phase and coherence estimations. (10.1109/JSTARS.2026.3722046)
  DOI : 10.1109/JSTARS.2026.3722046
• Module Learning With Errors and Structured Extrapolated Dihedral Cosets
  
  • Wen Weiqiang
  • Zheng Jinwei
  , 2026, 16802, pp.263-294. <div><p>The Module Learning With Errors (MLWE) problem is the fundamental hardness assumption underlying the key encapsulation and signature schemes ML-KEM and ML-DSA, which have been selected by NIST for post-quantum cryptography standardization. Understanding its quantum hardness is crucial for assessing the security of these standardized schemes. Inspired by the equivalence between LWE and Extrapolated Dihedral Cosets Problem (EDCP) in [Brakerski, Kirshanova, Stehlé and Wen, PKC 2018], we show that the MLWE problem is as hard as a structured variant of the EDCP, which we refer to as the Integer Polynomial Module EDCP (IP-M-EDCP). This extension from EDCP to IP-M-EDCP relies crucially on the algebraic structure of the ring underlying MLWE: the extrapolation depends not only on the noise rate, but also on the ring's degree. In fact, an IP-M-EDCP state forms a superposition over an exponential (in ring degree) number of possibilities. Our equivalence result holds for MLWE defined over power-of-two cyclotomic rings with constant module rank, a setting of particular relevance in cryptographic applications. Moreover, we present a reduction from IP-M-EDCP to EDCP. Therefore, to analyze the quantum hardness of MLWE, it may be advantageous to study IP-M-EDCP, which might be easier than EDCP.</p></div> (10.1007/978-3-032-35377-1_9)
  DOI : 10.1007/978-3-032-35377-1_9
• Equivariant Denoisers for Plug and Play Image Restoration
  
  • Renaud Marien
  • Guez Eliot
  • Leclaire Arthur
  • Papadakis Nicolas
  Journal of Mathematical Imaging and Vision, Springer Verlag, 2026, 68 (54). One key ingredient of image restoration is to define a realistic prior on clean images to complete the missing information in the observation. State-of-the-art restoration methods rely on a neural network to encode this prior. Typical image distributions are invariant to some set of transformations, such as rotations or flips. However, most deep architectures are not designed to represent an invariant image distribution. Recent works have proposed to overcome this difficulty by including equivariance properties within a Plug-and-Play paradigm. In this work, we propose two unified frameworks named Equivariant Regularization by Denoising (ERED) and Equivariant Plug-and-Play (EPnP) based on equivariant denoisers and stochastic optimization. We analyze the convergence of the proposed algorithms and discuss their practical benefit.
• On Pairwise Quantile Regression - Statistical Guarantees and Applications
  
  • Thérézien Romain
  • Clémençon Stéphan
  • Girard Fantin
  • El-Abdouni Hamza
  , 2026.
• Distributed Acoustic Sensing Over Deployed Telecommunication Fiber Networks: Opportunities and Challenges
  
  • Awwad Élie
  • Freire-Hermelo Maria
  • Prato Diane
  • Pruvost Pierre
  • Kumar Choudhury Pallab
  • Jaouën Yves
  • Gabet Renaud
  • Huang Heming
  , 2026. In this paper, we highlight opportunities and challenges of deploying Rayleigh-based distributed optical fiber sensing, also known as distributed acoustic sensing (DAS), over existing optical telecommunication networks. We review some of the prominent recent advances in this area and we shed light on future directions. First, we briefly review classifications of DAS systems and we focus on the advantages of using a continuous-wave constant-power DAS probe signal for inband coexistence between DAS and data transmission over the same fiber. Second, we emphasize on the important role of the mechanical coupling between the fiber, its hosting cable and their surrounding environment in the interpretation of the extracted strain information. Third, we show results on the estimation of distributed birefringence information using a polarization-diversity DAS interrogation. Finally, we study the impact of laser phase noise on continuous-wave constant-power DAS interrogation using phase-encoded probing sequences and attempt to mitigate it.
• Comparative Analysis of Spread-Spectrum Codes for Fibre-Optic Distributed Acoustic Sensing
  
  • Freire-Hermelo Maria
  • Awwad Élie
  , 2026. <div><p>This paper presents a comprehensive performance comparison of several coded sequences for distributed acoustic sensing systems through numerical simulations. Correlation analysis reveals that perfect autocorrelation sequences provide accurate channel estimation. Phase noise sensitivity varies significantly with frequency diversity, with high-diversity sequences maintaining lower estimation errors for phase and intensity despite increased laser linewidths. For a coded distributed acoustic sensing system with 10 Hz laser linewidth, high-diversity sequences achieve 2.7 × 10 -3 rad phase error compared to 5.3 × 10 -3 rad for low-diversity codes for a 1.5 km fibre length. The findings establish performance trade-offs between correlation quality, phase noise resilience, and practical implementation constraints for next-generation fibre-optic distributed acoustic sensing systems.</p></div>
• Theoretical comparison of independent-based samplers
  
  • Bertholom François
  • Douc Randal
  • Roueff François
  , 2026. In this work, we analyze the convergence properties of four independent-based Markov Chain Monte-Carlo samplers, the Importance Markov Chain and three algorithms based on multiple tries. We prove that the Multiple-Try Metropolis with Independent Sampling dominates Multiple-Try Metropolis with Independent Balancing and Iterated Sampling Importance Resampling in the Peskun sense, guaranteeing its theoretical superiority over these alternatives. Furthermore, we derive a general formula for obtaining exact uniform convergence rates of these algorithms, assuming that the density ratio between the target and the proposal is bounded. In the unbounded case, while none of the three algorithms based on multiple tries are geometrically ergodic, the Importance Markov Chain only requires a simple exponential moment condition to be geometrically ergodic, or a polynomial moment condition to ensure polynomial ergodicity.