Portail HAL des publications du LIRMM
Not a member yet
13279 research outputs found
Sort by
A Runtime Efficient Graph-Based Cell-Aware Model Generation for Structural SRAM Testing
No full textInternational audienceTesting advanced memories is essential for ensuring modern System-on-Chip quality. As transistor size continues to shrink, the probability of the occurrence of manufacturing defects increases, making conventional functional testing of SRAMs inadequate for achieving required Defect Parts per Million (DPPM). To address this issue, a novel structural testing approach using the Cell-Aware (CA) test methodology has been proposed in [1]. With this methodology, structural test patterns are obtained through an Automatic Test Pattern Generator (ATPG) by exploiting analog CA models (i.e., based on exhaustive analog simulations) for SRAM primary blocks. However, the generation of CA models through analog simulations is timeconsuming and technology dependent. A methodology, namely TrUnDeL proposed in [2], is used to accelerate the CA model generation, combining a switch-level graph-based solution and analog simulations. In this work, we propose an adaptation of TrUnDeL to generate the CA models, using only the switch-level graph-based simulations. TrUnDeL CA models are then used by the ATPG to generate structural test patterns. Through a validation flow, we demonstrate that the aforementioned patterns, generated without running any analog simulations, achieve nearly the same fault coverage as the test patterns generated using exhaustive analog simulations on an SRAM case study. The time required to generate CA models is drastically reduced with TrUnDeL, from 1 hour to 15 seconds for the considered case study
On Extracting Legal Arguments
No full textInternational audienceBuilding on the principles of case-based reasoning, we investigate the extraction of arguments from legal case databases. An argument is modeled as a set of factors that frequently support one party (plaintiff or defendant) over the other. The relevance of an argument is assessed by the number of cases that confirm it versus those that contradict it. Following established practices in data mining, we introduce a condensed representation of arguments called closed arguments, which capture the strongest form of support given the factors they contain. We develop propositional SAT-based encodings to enable the extraction of both arguments and closed arguments using SAT solvers. Additionally, we define a more compact condensed representation called maximal arguments, which eliminates redundancy by retaining only the most informative arguments with respect to given thresholds. We propose a level-wise algorithm that builds on our SAT-based approach for argument extraction. Preliminary experiments demonstrate the feasibility of our SAT-based mining methods
Accelerating first-order secure ML-KEM with masked SHA-3: cost, randomness, and security evaluation
No full textInternational audienceHash functions are fundamental for ensuring authenticity and integrity in digital communications. Their importance has grown with the advent of Post-Quantum Cryptography (PQC), as many newly standardized PQC algorithms, such as ML-KEM, heavily rely on hash functions for pseudo-random number generation. Among available standards, SHA-3, published by NIST in 2015, is the recommended choice for PQC applications. However, its adoption in embedded platforms remains limited, particularly in the presence of security threats such as side-channel attacks (SCAs).To mitigate SCAs, masking techniques are widely employed, but their implementation on SHA-3 accelerators is complex due to the non-linearity of the Keccak function. Domain-Oriented Masking (DOM) provides strong security guarantees but requires a significant amount of randomness, introducing additional implementation costs often overlooked in the literature. For instance, existing DOM implementations of Keccak demand 1600 bits of fresh randomness per clock cycle, raising practical concerns about randomness generation and deployment overhead. In this work, we present a hardware accelerator supporting all the SHA-3 functions with first-order masking countermeasure based on a DOM implementation of the Keccak core. For the generation of the randomness required by the DOM countermeasure, we implemented a randomness dispatcher based on Trivium and Bivium ciphers, and we evaluated the cost in terms of area of the circuit in FPGA and ASIC of the countermeasure and of the randomness generation. In addition, we assess its security through Test Vector Leakage Assessment on FPGA. Finally, the accelerator is integrated into a RISC-V-based 32-bit SoC and used to accelerate a first-order masked implementation of ML-KEM, achieving a speed-up ranging from 95x to 264x for the SHA-3-based functions of the KEM decapsulation
AutonoMoov: an Embedded Physio-Gaming Platform with Multi-level and Rhythmic Motor Interaction
No full textInternational audienceMotion-based video games, or physio-games, have shown promise in stimulating motor and cognitive functions among older adults and individuals with disabilities. However, their deployment remains limited due to the reliance on dedicated capture hardware and the lack of integrated support for both gross and fine motor interactions. In this work, we introduce AutonoMoov, a mobile physio-gaming platform that runs entirely on smartphones and tablets using their built-in RGB cameras. Leveraging MediaPipe for real-time markerless pose estimation, AutonoMoov supports a wide range of motor interactions, from large-scale limb movements to fine finger gestures, without external sensors. We also introduce Origins, a Unity library that bridges pose estimation with multi-level motion interaction and will include rhythmic control mechanisms to enhance engagement and embodiment. Our platform integrates several serious games and is designed for accessible, portable use in care homes, clinics, and domestic environments
Protein Function Prediction: Graph Neural Networks as Multimodal Aggregators of Sequences, Networks, and Ontologies
No full textNational audienceBackgroundProtein function prediction has evolved from sequence-based methods to deep learning-based multimodal approaches incorporating diverse data streams to better reflect biological complexity. Graph Neural Networks (GNNs) are increasingly used to integrate heterogeneous data due to their ability to handle different dimensionalities and scales. However, the contributions of various data modalities to prediction accuracy remain unclear.ResultsWe use a simplified deep learning framework with a single-layer attention-based GNN on the BeProf dataset to investigate the impact of different data modalities on protein function prediction. The results confirm GNNs as effective integrators of diverse data, including protein interaction networks, textual, sequence and structural information. Our findings highlight the central role of language modelembeddings and the benefits of integrating InterPro and Gene Ontology annotations. We show that structural information, while useful at the protein scale[1], is redundant with other data types in large-scale networks. Despite using a simple architecture, we reach first place on the BeProf benchmark’s Cellular Component subset, third and fifth respectively on the Biological Processes and Molecular Function subsets, outperforming popular methods such as DeepGoZero, DeepGraphGO or TALE.ConclusionThis study provides insights into the relative contributions of different data modalities, showing that sequence information remains foundational while interactome, structural, and ontology-based features offer avenues for improvement. We highlight the trade-offs and benefits of multimodal frameworks, providing guidelines for future innovations in computational biology and machinelearning for enhanced protein functional annotation
Prédiction des séquences culturales par apprentissage automatique : une comparaison entre grandes cultures et maraîchage diversifié
No full textInternational audienceCrop rotation is the practice of alternating different crops on the same plot for agronomic benefit. However, this practice is difficult because it depends on agronomic rules but also on experience in the field to know which rotations are beneficial, neutral or to be avoided. We propose a methodology for implementing a machine learning model for predicting crop sequences based on the farmer’s cropping history and context.La rotation des cultures est une pratique consistant à alterner différentes cultures sur une même parcelle pour des bénéfices agronomiques. Cependant, cette pratique est difficile car elle est dépendante de règles agronomiques mais aussi d'une expérience du terrain permettant de connaître les rotations bénéfiques, neutres ou à éviter. Nous proposons une méthodologie d'implémentation de modèle d'apprentissage machine pour la prédiction de séquence de culture basé sur l'historique mais aussi sur le contexte de culture de l'agriculteur
Towards Decentralized Health Data Platforms
No full textInternational audienceRecently approved European Health Data Space (EHDS) regulation is envisioned to promote the medical research and provide secure and privacy-preserving data access and management on the crossborder scale. However, the development and, more importantly, adaption of the supporting technology for such ambitious project is still ongoing. This issue is also complicated by varying legal frameworks between different member states. Our work presents a decentralized data platform approach for enforcing such multi-national legal acts as well as verification and auditing mechanism
Toward Adaptive MLOps : Variability Mapping and Modeling
No full textNational audienceMLOps knows a growing diversity of pipelines, characterized by varying structures, tools,and execution strategies. While this flexibility supports a range of use cases, it also intro-duces significant variability, which poses challenges for standardization and reuse. To ad-dress this, we explore Software Process line (SPrL) paradigm to manage variability acrossfamilies of pipelines. Our research investigates the landscape of variability in MLOpspipelines, using a Systematic Mapping Study of 21 primary papers, we categorize MLOpsvariability into 6 categories and derive modeling requirements to represent them. We thenpropose a modeling approach following those requirements
Tactile-based force estimation for interaction control with robot fingers
No full textInternational audienceFine dexterous manipulation requires reactive control based on rich sensing of manipulator-object interactions. Tactile sensing arrays provide rich contact information across the manipulator’s surface. However their implementation faces two main challenges: accurate force estimation across complex surfaces like robotic hands, and integration of these estimates into reactive control loops. We present a data-efficient calibration method that enables rapid, full-array force estimation across varying geometries, providing online feedback that accounts for non-linearities and deformation effects. Our force estimation model serves as feedback in an online closed-loop control system for interaction force tracking. The accuracy of our estimates is independently validated against measurements from a calibrated force-torque sensor. Using the Allegro Hand equipped with Xela uSkin sensors, we demonstrate precise force application through an admittance control loop running at 100Hz, achieving up to 0.12±0.08 [N] error margin—results that show promising potential for dexterous manipulation
Summer School FLOW 2025 : Introduction to the Semantic Web
No full textMasterOn May 21, 2025, during the FLOW2025 Summer School held at École Polytech Montpellier, this introductory course explored the foundations and key technologies of the Semantic Web. The session began by highlighting the limitations of the current Web, which is primarily designed for human consumption, and presented the Semantic Web as an extension that allows machines to interpret, structure, and connect web data using formal standards. The course introduced core concepts such as RDF for representing information as graphs of triples, RDFS for defining semantic relationships between data elements, and SPARQL for querying graph-based knowledge bases. Through demonstrations and interactive examples, participants were introduced to the logic and tools that underpin semantic interoperability. The session concluded by emphasizing the importance of these technologies in implementing FAIR data principles and supporting open science.Le 21 mai 2025, dans le cadre de l’école d’été FLOW2025 organisée à l’École Polytech Montpellier, ce cours d’introduction a porté sur les fondements et les technologies clés du Web sémantique. La session a débuté par une présentation des limites du Web actuel, principalement orienté vers les usages humains, avant d’introduire le Web sémantique comme une extension permettant aux machines d’interpréter, structurer et relier les données du Web à l’aide de standards formels. Les concepts essentiels tels que RDF pour la représentation en triplets, RDFS pour la structuration des relations sémantiques, et SPARQL pour l’interrogation des graphes de connaissances ont été abordés de manière progressive. Des démonstrations et des exemples interactifs ont permis aux participant·e·s de se familiariser avec les logiques et outils du Web sémantique. La session s’est conclue sur l’importance de ces technologies pour l’interopérabilité des données, la mise en œuvre des principes FAIR et le soutien à la science ouverte