Portail HAL des publications du LIRMM
Not a member yet
13279 research outputs found
Sort by
Designing Compatible Analog Circuits for Equilibrium Propagation: Implementations Using The Adjoint Method and Reciprocity Principles
No full textInternational audienceEquilibrium Propagation (EP) offers an energyefficient alternative to backpropagation for training energy-based models (EBM) in analog neural networks, making it a promising approach for on-device learning in Edge AI applications. However, practical hardware implementations of EP are hindered by stringent circuit requirements and challenges in computing loss gradients for common functions like crossentropy without resorting to expensive data converters. In this paper, we address these challenges by identifying the necessary conditions that an analog circuit must satisfy to implement EP, leveraging the adjoint method and the reciprocity principle of analog circuits, and developing two novel circuit architectures that fully realize EP in hardware. The first architecture employs the sparsemax activation function, offering a practical means to implement a crossentropy-like loss function while circumventing the complexities associated with the softmax function. The second architecture introduces a simpler topology that aligns with analog processing principles by enforcing output values to reside in a simplex, eliminating the need for calculating probabilities explicitly. We validate our designs on two benchmark datasets, demonstrating that our circuits can effectively and efficiently train analog neural networks using EP within practical hardware constraints
Combinatoire des chevauchements d'une paire de mots
No full textAdditional file: Diaporama of the Presentation made at COCOON by P. WangInternational audienceEine Korrelation ist ein binärer Vektor, der alle möglichen Positionen von Überschneidungen zweier Wörter kodiert, wobei eine Überschneidung für ein geordnetes Wortpaar (u, v) auftritt, wenn ein Suffix von u mit einem Präfix von v übereinstimmt. Da mehrere Paare dieselbe Korrelation aufweisen können, ist es relevant zu zählen, wie viele Wortpaare dieselbe Korrelation teilen, abhängig von der Alphabetgröße und der Wortlänge n. Wir zeigen Rekursionen auf, um die Anzahl solcher Paare – die als Populationsgröße bezeichnet wird – für jede Korrelation zu berechnen; dazu nutzen wir eine Beziehung zwischen Überschneidungen zweier Wörter und der Selbstüberschneidung eines Wortes. Dieser Satz ermöglicht es uns, die Anzahl der Paare mit der längsten Überschneidung einer gegebenen Länge zu berechnen und damit zwei offene Fragen zu lösen, die Gabric 2022 aufgeworfen hat. Schließlich liefern wir auch Grenzen für das asymptotische Populationsverhältnis jeder Korrelation. Angesichts der Bedeutung von Wortüberlappungen in Bereichen wie der Kombinatorik auf Wörtern, der Bioinformatik und der digitalen Kommunikation können unsere Ergebnisse die Analyse von Algorithmen für die Zeichenfolgenverarbeitung, das Codedesign oder die Genomassemblierung erleichtern.A correlation is a binary vector that encodes all possible positions of overlaps of two words, where an overlap for an ordered pair of words (u, v) occurs if a suffix of u matches a prefix of v. As multiple pairs can have the same correlation, it is relevant to count how many pairs of words share the same correlation, depending on the alphabet size and word length n. We exhibit recurrences to compute the number of such pairs -which is termed population size -for any correlation; for this, we exploit a relationship between overlaps of two words and self-overlap of one word. This theorem allows us to compute the number of pairs with the longest overlap of a given length, solving two open questions Gabric raised in 2022. Finally, we also provide bounds for the asymptotic population ratio of any correlation. Given the importance of word overlaps in areas like combinatorics on words, bioinformatics, and digital communication, our results may ease analyses of algorithms for string processing, code design, or genome assembly.Une corrélation est un vecteur binaire qui encode toutes les positions possibles de chevauchement entre deux mots, où un chevauchement pour une paire ordonnée de mots (u, v) se produit si un suffixe de u correspond à un préfixe de v. Étant donné que plusieurs paires peuvent avoir la même corrélation, il est pertinent de compter combien de paires de mots partagent la même corrélation, en fonction de la taille de l'alphabet et de la longueur des mots n. Nous présentons des récurrences pour calculer le nombre de ces paires - appelé taille de population - pour toute corrélation ; pour cela, nous exploitons une relation entre les chevauchements de deux mots et l'auto-chevauchement d'un mot. Ce théorème nous permet de calculer le nombre de paires ayant le plus long chevauchement d'une longueur donnée, résolvant ainsi deux questions ouvertes soulevées par Gabric en 2022. Enfin, nous fournissons également des limites pour le rapport asymptotique de population de toute corrélation. Compte tenu de l'importance des chevauchements de mots dans des domaines tels que la combinatoire sur les mots, la bio-informatique et la communication numérique, nos résultats pourraient faciliter l'analyse des algorithmes pour le traitement des chaînes de caractères, la conception de codes ou l'assemblage de génomes
Blood donors as sentinels for genomic surveillance of West Nile virus in Germany using a sensitive amplicon-based sequencing approach
No full textInternational audienceBackground: West Nile virus (WNV) has emerged as a public health concern in Germany since its first detection in 2018, with evidence of expanding geographic spread. Genomic surveillance is critical for tracking viral evolution, identifying introductions, and monitoring local transmission. However, genome recovery from lowviremia samples such as those obtained through blood donor screening remains technically challenging. Aim: To develop and validate a sensitive amplicon-based sequencing protocol optimized for WNV lineage 2 and apply it to low-titer samples to support genomic surveillance in Germany. Methods: A novel primer scheme was designed for WNV lineage 2 and applied to 43 nucleic acid testing (NAT)-positive blood donor samples collected between 2020 and 2024. Amplicon-based sequencing performance was benchmarked against metagenomic next-generation sequencing (mNGS). Recovered genomes were subjected to phylogenomic analysis to assess viral diversity and transmission dynamics. Results: The amplicon protocol enabled genome recovery (> 70% coverage) from samples with viral loads as low as ∼10¹ RNA copies/µL, outperforming metagenomic NGS (mNGS). Of the 43 samples, 30 yielded complete or nearcomplete genomes. Six distinct WNV subclades (2A-2F), including German strains, were identified, indicating multiple introductions into Germany from Central Europe. Subclade 2F emerged as the dominant and widely distributed group. Berlin, Brandenburg, Saxony, and Saxony-Anhalt were identified as persistent transmission hubs. Conclusion: This study highlights blood donors as valuable sentinels for WNV genomic surveillance. The validated amplicon-based sequencing approach enables sensitive, scalable genome recovery from lowviremia samples, and when integrated with routine blood donor screening, provides a robust framework for early detection, transmission dynamics, and public health preparedness.</div
Leveraging gem5 and Machine Learning for End-to-End Detection of Cache-based Side-Channel Attack Patterns
No full textInternational audienceSide-channel attacks (SCAs) exploit physical leakage vectors, including cache states, timing variations, and power consumption in hardware microarchitectures to compromise computational security. This paper introduces an end-to-end, simulation-driven framework that integrates gem5's cycle-accurate architectural models with unsupervised machine learning to automate SCA detection. We simulate Spectre (V1/V2), Prime+Probe, and Flush+Reload attack pattern workloads to generate fine-grained execution traces encompassing BTB access sequences, memory latency distributions, and pipeline stall events. Temporal feature engineering extracts discriminative signatures through branch predictor entropy calculations and miss sequence autocorrelation, while dimensionality reduction via t-SNE optimizes the feature space. These preprocessed traces train an ensemble of Isolation Forest, Variational Autoencoders, and HDBSCAN models to identify attack patterns without predefined templates. Experimental validation demonstrates 0.92 precision and 0.88 recall for cache-based SCAs (Prime+Probe variants), representing a 92% F1-score improvement over SVM baselines. SHAP-based feature attribution reveals BTB-miss run-lengths and memory controller contention queues as critical attack patterns. This pipeline automatically categorize interpretable attack patterns if they are risky for the system, enabling proactive identification of safe and unsafe instructions for secure hardware through gem5's reconfigurable memory hierarchy and cache partitioning mechanisms
Cost-Optimized Double-Node-Upset-Recovery Latch Designs With Aging Mitigation and Algorithm-Based Verification for Long-Term Robustness Enhancement
No full textInternational audienceWith the continuous advancement of CMOS technologies, soft errors, such as single-node upset (SNU) and double-node upset (DNU), caused by radiation in nanoscale integrated circuits, are becoming increasingly prominent. Meanwhile, transistor aging mitigation is indispensable for long-term robustness enhancement. First, to reduce the impact of radiation on circuits, we propose a novel DNU-recovery latch with low cost, namely, DURLC, only consisting of four dual-input C-elements (CEs) and four clock-gated input-split inverters for the storage of values. Second, we propose a DNU-recovery latch with moderate cost, namely, DURMC, based on seven CEs and four inverters, for convenience to optimize the latch to alleviate aging. The proposed DNU-recovery latch with mitigated aging is called DURMA. The latch employs a high-speed path to reduce delay without sacrificing performance when mitigating aging issues. Finally, we propose an algorithm-based verification method to validate the DNU recovery of the proposed latches. The simulation results show that, compared with the state-of-the-art robust latches, the proposed latches have the advantages of DNU recovery with moderate and even low cost, and meanwhile, aging is effectively mitigated for the DURMA latch
Design of Nonvolatile and Multinode-Upset Recoverable Latches Based on Magnetic Tunnel Junction and CMOS
No full textInternational audienceSpintronic devices, such as magnetic tunnel junctions (MTJs), are promising for space applications due to their radiation hardness and nonvolatility. However, as semiconductor technology advances, CMOS peripheral circuits are becoming vulnerable to double node upset (DNU) as well as triple node upset (TNU). This brief proposes two nonvolatile and robust latch designs primarily composed of MTJs and C-elements (CEs). Both designs offer nonvolatility and self-recovery from multiple-node upsets. Simulation results demonstrate that the proposed latches provide nonvolatility and complete protection against multiplenode upsets with balanced overhead
Identification automatique de la charge de pollen des abeilles et application aux études de terrain
No full textInternational audienceIn a changing world, it is crucial to characterise communities and their evolution over time. Because social insect pollinators forage on flowering plants around the colony, the nest potentially contains important information about the pollinated plants such as species identity and plant phenology. In this paper, we introduce new approaches to assess plant composition in a Mediterranean summer plant community from pollen foraged by honeybees. We leveraged the autofluorescence properties of the pollen load to classify plant species, both using a UV/Vis spectrophotometer in the laboratory and a dedicated prototype ‘pollen analyser’ adapted to field studies. Our results demonstrate that data collected from fluorescent spectra and pollen analyser measurements of pollen load from 14 plant species are specific enough to distinguish plant species. When combined with machine learning techniques, particularly the Support Vector Machine classifier, these approaches provide powerful methods to automatically identify species from fluorescence measurements of pollen load. Overall, our study shows that analysing the autofluorescence of honeybee pollen load enables the precise identification of their floral origins, paving the way for a real‐time, spatially distributed observatory of flowering plants to monitor species identity, flowering phenology and long‐term ecological dynamics.Dans un monde en évolution, il est crucial de caractériser les communautés et leur évolution au fil du temps. Étant donné que les insectes pollinisateurs sociaux se nourrissent de plantes à fleurs autour de la colonie, le nid contient potentiellement des informations importantes sur les plantes pollinisées, telles que l'identité de l'espèce et la phénologie des plantes. Dans cet article, nous introduisons de nouvelles approches pour évaluer la composition végétale dans une communauté végétale estivale méditerranéenne à partir du pollen récolté par les abeilles. Nous avons exploité les propriétés d’autofluorescence de la charge pollinique pour classer les espèces végétales, à la fois en utilisant un spectrophotomètre UV/Vis en laboratoire et un prototype dédié « analyseur de pollen » adapté aux études sur le terrain. Nos résultats démontrent que les données collectées à partir des spectres de fluorescence et des mesures de la charge pollinique de 14 espèces végétales par un analyseur de pollen sont suffisamment spécifiques pour distinguer les espèces végétales. Lorsqu’elles sont combinées à des techniques d’apprentissage automatique, en particulier le classificateur Support Vector Machine, ces approches fournissent des méthodes puissantes pour identifier automatiquement les espèces à partir de mesures de fluorescence de la charge pollinique. Dans l’ensemble, notre étude montre que l’analyse de l’autofluorescence de la charge de pollen des abeilles permet l’identification précise de leurs origines florales, ouvrant la voie à un observatoire en temps réel et spatialement distribué des plantes à fleurs pour surveiller l’identité des espèces, la phénologie de la floraison et la dynamique écologique à long terme
PhenoTrack3D: an automatic high-throughput phenotyping pipeline to track maize organs over time
No full textBackground High-throughput phenotyping platforms allow the study of the form and function of a large number of genotypes subjected to different growing conditions (GxE). A number of image acquisition and processing pipelines have been developed to automate this process, for micro-plots in the field and for individual plants in controlled conditions. Capturing shoot development requires extracting from images both the evolution of the 3D plant architecture as a whole, and a temporal tracking of the growth of its organs. Results We propose PhenoTrack3D, a new pipeline to extract a 3D+t reconstruction of maize at organ level from plant images. It allows the study of plant architecture and individual organ development over time during the entire growth cycle. PhenoTrack3D improves a former method limited to 3D reconstruction at a single time point [Artzet et al ., 2019] by (i) a novel stem detection method based on deep-learning and (ii) a new and original multiple sequence alignment method to perform the temporal tracking of ligulated leaves. Our method exploits both the consistent geometry of ligulated leaves over time and the unambiguous topology of the stem axis. Growing leaves are tracked afterwards with a distance-based approach. This pipeline is validated on a challenging dataset of 60 maize hybrids imaged daily from emergence to maturity in the PhenoArch platform (ca. 250,000 images). Stem tip was precisely detected over time (RMSE < 2.1cm). 97.7% and 85.3% of ligulated and growing leaves respectively were assigned to the correct rank after tracking, on 30 plants x 43 dates. The pipeline allowed to extract various development and architecture traits at organ level, with good correlation to manual observations overall, on random subsets of 10 to 355 plants. Conclusions We developed a novel phenotyping method based on sequence alignment and deep-learning. It allows to characterise automatically and at a high-throughput the development of maize architecture at organ level. It has been validated for hundreds of plants during the entire development cycle, showing its applicability to the GxE analyses of large maize datasets
Assessing the Nature of Human Brain‐Derived Extracellular Vesicles on Synaptic Activity Via the Development of an Air‐liquid Microfluidic Platform
No full textInternational audienceBrain‐Derived Extracellular Vesicles (BDEVs) have been associated with important roles in functional neuron networks. However, the various models that have been used to study these roles fail to account for all the specificities of the human brain. This study presents a microfluidic platform capable of injecting and/or collecting BDEVs from Organotypic culture of Post‐mortem Adult human Brain explants (OPAB) cultured at the air‐liquid interface, while measuring electrical activity in real‐time on 3D‐microelectrode arrays (MEA). The platform design and custom‐made program to control the system allows the automatic collection of BDEVs over days. Mass spectrometry analyses highlight that BDEVs are significantly enriched with synaptic proteins, such as Neural cell adhesion molecule, Syntaxin‐1A, and Synaptopodin, known to regulate synaptic plasticity. Using the MEA‐embedded air‐liquid microfluidic platform, it is shown that BDEVs injection on OPAB induces a significant decrease of local field potential compared to mock conditions, in particular for high frequency oscillations. Finally, a machine learning framework, experimentally validated, revealed that the co‐treatment of OPAB with BDEVs and GW4869, an inhibitor of exosome production, can counteract electrical perturbations induced by BDEVs alone. Together, this work provides innovative methodological developments, that contributed to reveal the diverse biological functions of BDEVs on neural activity
Advent of code, jour 17 (le retour)
No full textNational audienceRappelez-vous, je vous avais laissé dans un état de tension insupportable à l’issue de mon précédent article sur la première partie de l’énigme du 17e jour de l’Advent of Code (ce fameux mini calendrier de l’avent du code). Voici le moment de vous révéler ce qui nous attendait dans la seconde partie