Portail HAL des publications du LIRMM
Not a member yet
13279 research outputs found
Sort by
Injection de fautes par impulsion dans le susbstrat : modélisation
No full textFault injection techniques have been extensively developed and studied in the past decades. Variousapproaches have emerged, using different physical quantities' manipulation to disturb integrated circuitsbehavior, such as electromagnetic fields to create parasitic currents in the IC targets, also knownas electromagnetic fault injection. Additionally, light emission with the use of laser, leveraging thesilicon photoelectric effect, also known as laser fault injection. Then, simple power supply voltageor clock signal manipulations, enabling to run integrated circuits outside their specifications, thusprovoking exploitable unexpected behavior, known as glitch fault injection. Eventually, biasing thesubstrate of integrated circuits with the help of fast high-voltage pulses, known as body biasing injection,not to cite them all. Among these methods, laser fault injection, electromagnetic fault injectionor glitch fault injection were extensively studied. Therefore, countermeasures allowing to protectand secure integrated circuits were proposed for most of them. However, body biasing injection didnot get this attention. Consequently, no specific countermeasure or understanding of this techniquewere highlighted.In this context, my thesis' work brings some answers and insights concerning body biasing injection.I propose various enhancements of the state-of-the-art way of practicing body biasing injection,thus enabling more repeatable and more reliable experiments. Subsequently, I demonstrate the feasibility,thanks to body biasing injection, of a differential fault attack relying on a constraining single-bitfault model. Then, I introduce a new modeling and simulation flow dedicated to the study of bodybiasing injection. This methodology, based on previous works on electromagnetic fault injection,allows understanding the effect of body biasing injection on integrated circuits, going from electriccharge behavior, to logic gates disturbance and a fault model. Eventually, I present the study ofthe interest in thinning the substrate of integrated circuits for the practice of body biasing injection,illustrating the consequences of such a practice on the efficiency of this fault injection method.Ces dernières décennies, les méthodes d'injection de fautes ont été étudiées de manière approfondie.De nombreuses approches ont vu le jour, toutes utilisant diverses grandeurs physiques dans le butde perturber le comportement des circuits intégrés ciblés. On peut notamment identifier les champsélectromagnétiques permettant de créer des courants parasites dans les circuits, connu sous le nomd'injection de fautes par impulsion électromagnétique. De plus, il existe une technique utilisant lesémissions de photons produites grâce au laser, tirant profit de l'effet photoélectrique du silicium,connues sous le nom d'injection de fautes par impulsion laser. Ensuite, on trouve des méthodesplus élémentaires, tirant parti de la perturbation des signaux d'alimentation ou d'horloge. Celle-cicréent des comportements indésirables exploitables, communément appelés injection de fautespar "glitch". Enfin, on peut identifier la polarisation transitoire du substrat des circuits intégrés,aussi connue sous le nom d'injection de fautes par impulsion dans le substrat. Parmi ces méthodes,l'injection par impulsions laser, par impulsions électromagnétiques et par glitch ont été étudiées en détail. En revanche, l'injection de fautes par impulsion dans le substrat n'a pas eu ce traitement. Parconséquent, aucune contremesure dédiée n'a encore été proposée pour prévenir l'utilisation de cettetechnique.Dans ce contexte, mon travail de thèse de doctorat vise à apporter des éléments de réponse concernantl'injection de fautes par impulsion dans le substrat. Dans un premier temps, je propose desaméliorations des plateformes préexistantes, permettant de réaliser des expérimentations plus reproductibleset plus fiables. Ensuite, je démontre la faisabilité d'une attaque par faute différentiellequi s'appuie sur un critère contraignant. Ultérieurement, je présente une nouvelle méthodologie demodélisation et de simulation de l'injection de fautes par impulsion dans le substrat. Cette méthode,fondée sur de précédents travaux réalisés concernant l'injection de fautes par impulsion électromagnétique, permet d'avoir une compréhension sur les comportements internes d'un circuit soumis àl'injection par impulsion dans le substrat. Enfin, j'étudie l'intérêt d'affiner le substrat des circuits intégrésdestinés à être utilisés avec l'injection par impulsion dans le substrat, ce qui permet de prévoirles conséquences de cette pratique sur la méthode d'injection de fautes
TRAVERSéES: Territorial levers and transition pathways for reducing pesticide use
No full textThis issue deals with innovation in the context of changing societal demands and agri-food processing companies, and the relationships between foodtech players. It follows on from the Carrefour de l'innovation agronomique Foodtech, Innover, pourquoi et comment? held at AgroParisTech on 30 November 2023.International audienceThe TRAVERSéES project aims to identify trajectories for reducing pesticide use by leveraging various territorial tools. To achieve this, we first conducted an analysis of ecological, economic, social, institutional, and individual factors that influence changes in phytosanitary practices. These insights informed the development of a socio-ecosystem model designed to simulate agricultural practice trajectories across territories, which subsequently served as a tool for prospective analysis with stakeholders in the Barrois region (Grand-Est). The project employed a range of methodologies and engaged in a transdisciplinary partnership. This approach led to the emergence of multiple, innovative proposals for territorial transition levers and highlighted the diverse factors considered by farmers, along with varying degrees of sensitivity to these influences
Overview of PlantCLEF 2024: multi-species plant identification in vegetation plot images
No full textSource Agritrop Cirad (https://agritrop.cirad.fr/613025/) * Autres projets (id;sigle;titre): 101060693;GUARDEN;(EU) safeGUARDing biodivErsity aNd critical ecosystem services across sectors and scales// 101060639;MAMBO;(EU) Modern Approaches to the Monitoring of BiOdiversity//International audiencePlot images are essential for ecological studies, enabling standardized sampling, biodiversity assessment, longterm monitoring and remote, large-scale surveys. Plot images are typically fifty centimetres or one square meter in size, and botanists meticulously identify all the species found there. The integration of AI could significantly improve the efficiency of specialists, helping them to extend the scope and coverage of ecological studies. To evaluate advances in this regard, the PlantCLEF 2024 challenge leverages a new test set of thousands of multi-label images annotated by experts and covering over 800 species. In addition, it provides a large training set of 1.7 million individual plant images as well as state-of-the-art vision transformer models pre-trained on this data. The task is evaluated as a (weakly-labeled) multi-label classification task where the aim is to predict all the plant species present on a high-resolution plot image (using the single-label training data). In this paper, we provide an detailed description of the data, the evaluation methodology, the methods and models employed by the participants and the results achieved
Models and Methods for Biological Evolution: Mathematical Models and Algorithms to Study Evolution
No full textInternational audienceBiological evolution is the phenomenon concerning how species are born, are transformed or disappear over time. Its study relies on sophisticated methods that involve both mathematical modeling of the biological processes at play and the design of efficient algorithms to fit these models to genetic and morphological data.Models and Methods for Biological Evolution outlines the main methods to study evolution and provides a broad overview illustrating the variety of formal approaches used, notably including combinatorial optimization, stochastic models and statistical inference techniques.Some of the most relevant applications of these methods are detailed, concerning, for example, the study of migratory events of ancient human populations or the progression of epidemics.This book should thus be of interest to applied mathematicians interested in central problems in biology, and to biologists eager to get a deeper understanding of widely used techniques of evolutionary data analysis
Ambiguïté des étiquettes en classification et retours d'experts
No full textWhile classification datasets are composed of more and more data, the need for human expertise to label them is still present. Crowdsourcing platforms are a way to gather expert feedback at a low cost. However, the quality of these labels is not always guaranteed. In this thesis, we focus on the problem of label ambiguity in crowdsourcing. Label ambiguity has mostly two sources: the worker's ability and the task's difficulty. We first present a new indicator, the mathrm{WAUM} (Weighted Area Under the Magin), to detect ambiguous tasks given to workers. Based on the existing mathrm{AUM} in the classical supervised setting, this lets us explore large datasets while focusing on tasks that might require more relevant expertise or should be discarded from the actual dataset. We then present a new open-source texttt{python} library, PeerAnnot, that we developed to handle crowdsourced datasets in image classification. We created a benchmark in the Benchopt library to evaluate our label aggregation strategies for more reproducible results. Finally, we present a case study on the Pl@ntNet dataset, where we evaluate the current state of the platform's label aggregation strategy and propose ways to improve it. This setting with a large number of tasks, experts and classes is highly challenging for current crowdsourcing aggregation strategies. We report consistently better performance against competitors and propose a new aggregation strategy that could be used in the future to improve the quality of the Pl@ntNet dataset. We also release this large dataset of expert feedback that could be used to improve the quality of the current aggregation methods and provide a new benchmark.Alors que les jeux de données de classification sont composés d'un nombre croissant de données, le besoin d'expertise humaine pour les étiqueter est toujours présent. Les plateformes de crowdsourcing sont un moyen de recueillir les commentaires d'experts à faible coût. Cependant, la qualité de ces étiquettes n'est pas toujours garantie. Dans cette thèse, nous nous concentrons sur le problème de l'ambiguïté des étiquettes dans le crowdsourcing. L'ambiguïté des étiquettes a principalement deux sources : la capacité du travailleur et la difficulté de la tâche. Nous présentons tout d'abord un nouvel indicateur, le mathrm{WAUM} (Weighted Area Under the Magin), pour détecter les tâches ambiguës confiées aux travailleurs. Basé sur le mathrm{AUM} existant dans le cadre supervisé classique, il nous permet d'explorer de grands jeux de données tout en nous concentrant sur les tâches qui pourraient nécessiter une expertise plus pertinente ou qui devraient être éliminées du jeu de données actuel. Nous présentons ensuite une nouvelle bibliothèque texttt{python} open-source, PeerAnnot, développée pour traiter les jeux de données crowdsourcées dans la classification d'images. Nous avons créé un benchmark dans la bibliothèque Benchopt pour évaluer nos stratégies d'agrégation d'étiquettes afin d'obtenir des résultats reproductibles facilement. Enfin, nous présentons une étude de cas sur l'ensemble de données Pl@ntNet, où nous évaluons l'état actuel de la stratégie d'agrégation d'étiquettes de la plateforme et proposons des moyens de l'améliorer. Ce contexte avec un grand nombre de tâches, d'experts et de classes est très difficile pour les stratégies d'agrégation de crowdsourcing actuelles. Nous faisons état de performances constamment supérieures à celles de nos concurrents et proposons une nouvelle stratégie d'agrégation qui pourrait être utilisée à l'avenir pour améliorer la qualité de l'ensemble de données Pl@ntNet. Nous publions également en plus de ce grand jeu de données, des annotations d'experts qui pourraientt être utilisées pour améliorer la qualité des méthodes d'agrégation actuelles et fournir un nouveau point de référence
Online correction of task registration and robot models from user input
No full textInternational audienceIn application domains such as surgical robotics, fully autonomous control remains a long-term ambition and the systems are mostly teleoperated. In this article, the presence of an operator in-the-loop is exploited to perform the online registration of an initially inaccurate haptic guidance and the calibration of robot kinematic models using operator's intention instead of relying on exteroceptive sensors. This is used to improve online haptic guidance in the context of shared control, or to progress towards automatic task completion after an initial learning phase. The method presented in this paper is based on an optimization in the task space to minimize the errors between the executed and desired trajectories, both estimated from models. This approach is particularly relevant when the execution of a planned task would suffer from errors that exteroceptive measurements could not fully correct, because of sensor inaccuracy or unavailability. A user study realized for a drawing task is detailed to illustrate that initially inaccurate task registration and robot models can be corrected from user inputs only. The results show that the proposed algorithm can learn the correct models, which in turns significantly improves the quality of the haptic guidance and decreases path deviations during the teleoperated task
Transactions on Large-Scale Data- and Knowledge-Centered Systems; Vol. LVI, LNCS 14790. Special Issue on Data Management - Principles, Technologies, and Applications
No full textFeMPIM: A FeFET-Based Multifunctional Processing-in-Memory Cell
No full textInternational audienceThe Von-Neumann memory wall bottleneck that keeps expanding is mainly caused by the frequent data transfer between the main memory and the processor. The Processing in-Memory (PiM) capabilities of emerging nonvolatile devices have the potential to partially alleviate the memory wall problem. In this paper, we use the ferroelectric field-effect transistor (FeFET), one of the emerging nonvolatile devices, to design a multifunctional processing in-memory cell, namely FeMPIM. It can perform multiple logic operations in computing mode as well as content searching in ternary content-addressable memory (TCAM) mode. Simulation results demonstrate the multifunctional capability of the proposed FeMPIM as well as its moderate overhead when compared with the complementary metal-oxide-semiconductor (CMOS) based and the existing FeFET-based devices
Rule-aware Datalog Fact Explanation Using Group-SAT Solver
No full textInternational audienceOne of the major benefits of symbolic AI is explainability. When new knowledge is obtained via a reasoning process, it is possible to determine precisely all elements of the knowledge base that yield this knowledge. Typically, one would use a SAT solver to compute the explanations. However, SAT-solving is computationally expensive, and as the knowledge base grows, the time required increases exponentially. This work presents a method for filtering a datalog knowledge base to optimise the time used by a SAT solver. This is achieved by creating a hypergraph representing the grounded knowledge base and pruning the nodes that are not reachable from the fact that we want to explain. This approach proves to be time-effective. Interestingly, one additional benefit of using this hypergraph is that it is possible to encode more information about the rules used in the reasoning process. By using an off-the-shelf group-SAT solver, this extra information allows us to find specific explanations that would be missed if we only considered facts
Killing a Vortex
No full textInternational audienceThe Graph Minors Structure Theorem of Robertson and Seymour asserts that, for every graph H, every H-minor-free graph can be obtained by clique-sums of "almost embeddable" graphs. Here a graph is "almost embeddable" if it can be obtained from a graph of bounded Eulergenus by pasting graphs of bounded pathwidth in an "orderly fashion" into a bounded number of faces, called the vortices, and then adding a bounded number of additional vertices, called apices, with arbitrary neighborhoods. Our main result is a full classification of all graphs H for which the use of vortices in the theorem above can be avoided. To this end we identify a (parametric) graph and prove that all -minor-free graphs can be obtained by clique-sums of graphs embeddable in a surface of bounded Euler-genus after deleting a bounded number of vertices. We show that this result is tight in the sense that the appearance of vortices cannot be avoided for H-minor-free graphs, whenever H is not a minor of for some t ∈ N.Using our new structure theorem, we design an algorithm that, given an -minor-free graph G, computes the generating function of all perfect matchings of G in polynomial time. Our results, combined with known complexity results, imply a complete characterization of minorclosed graph classes where the number of perfect matchings is polynomially computable: They are exactly those graph classes that do not contain every as a minor. This provides a sharp complexity dichotomy for the problem of counting perfect matchings in minor-closed classes