INRIA a CCSD electronic archive server
Not a member yet
122212 research outputs found
Sort by
Plant tolerance is explained by resource-based plant--nematode interactions
No full textPlant--parasitic nematodes are responsible for significant economic losses worldwide, affecting a wide range of crops. They infect and divert resources and impair plant growth. While resistance aims to limit parasite burden, tolerance is the ability of a plant to maintain its yield despite infection, which offers a complementary and potentially more sustainable strategy. Understanding how tolerance arises and operates requires mechanistic insights into host--parasite interactions and the constraints that shape them. In this study, we develop a mathematical model describing the coupled dynamics of plant growth, internal resource allocation, and nematode population processes. The model incorporates key biological mechanisms, including resource-dependent plant development, nematode infection, reproduction and feedbacks between resource limitation and parasitism. We analytically derive the basic reproduction number and identify equilibrium states, including the system extinction, healthy plantation and coexistence. Our analysis shows that coexistence strongly depends on plant resources at the time of infection, shedding light on conditions under which tolerance can emerge. Numerical simulations further explore how variations in both plant and nematode parameters influence the tolerance over a cropping season. We highlight the dual role of resource production: high production rates not only boost plant growth but also enhance nematode proliferation. There is hence a compromise to be found between favoring bigger and more productive plants, versus limiting yield losses in case of nematode infestation. This compromise is shaped by the nematode virulence and plant quantitative resistance
Optimal Best Arm Identification under Differential Privacy
No full textInternational audienceBest Arm Identification (BAI) algorithms are deployed in data-sensitive applications, such as adaptive clinical trials or user studies. Driven by the privacy concerns of these applications, we study the problem of fixed-confidence BAI under global Differential Privacy (DP) for Bernoulli distributions. While numerous asymptotically optimal BAI algorithms exist in the non-private setting, a significant gap remains between the best lower and upper bounds in the global DP setting. This work reduces this gap to a small multiplicative constant, for any privacy budget ϵ. First, we provide a tighter lower bound on the expected sample complexity of any δ-correct and ϵ-global DP strategy. Our lower bound replaces the Kullback-Leibler (KL) divergence in the transportation cost used by the non-private characteristic time with a new information-theoretic quantity that optimally trades off between the KL divergence and the Total Variation distance scaled by ϵ. Second, we introduce a stopping rule based on these transportation costs and a private estimator of the means computed using an arm-dependent geometric batching. En route to proving the correctness of our stopping rule, we derive concentration results of independent interest for the Laplace distribution and for the sum of Bernoulli and Laplace distributions. Third, we propose a Top Two sampling rule based on these transportation costs. For any budget ϵ, we show an asymptotic upper bound on its expected sample complexity that matches our lower bound to a multiplicative constant smaller than 8. Our algorithm outperforms existing δ-correct and ϵ-global DP BAI algorithms for different values of ϵ
Polarization MultiFocus Microscopy for volumetric super-resolution and orientation imaging of biofilaments
No full textInternational audienceAccessing molecular orientation in single molecule localization microscopy (SMLM) offers valuable insights into molecular ordering and organization in biological structures. Conventional single-molecule orientation-localization microscopy (SMOLM) methods typically rely on either engineering the microscope's point-spread function (PSF) to encode the orientation information or on polarization resolved detection. While PSF engineering enables detailed orientation analysis, it often requires complex computational analysis and suffers from reduced performance in dense cellular environments due to PSF spreading and overlap. In contrast, polarization-based approaches are easier to implement and are more fit when imaging dense samples but are unable to retrieve the axial information of single molecules. To overcome this limitation, we introduce the Polarization MultiFocus Microscope (PolMFM), a novel method for simultaneously retrieving the orientation and 3D position of single molecules. PolMFM combines the orientation measurement capabilities of a 4-polarization splitting scheme with a 3-planes multifocus microscope (MFM) enabling the reconstruction of molecular 2D orientation, wobble, and axial localization in a single acquisition. Through simulations, we demonstrate that PolMFM accurately recovers both orientation and 3D position, despite PSF defocusing. Experimental validation with reference samples shows that PolMFM matches the orientation precision of 4-Polar STORM, while uniquely adding axial information.We demonstrate the power of PolMFM by resolving the orientation and 3D positions of molecules in actin filaments in fixed cells, and by revealing that chromatin in crickets undergoes major reorganization and increased ordering during spermiogenesis. These findings highlight the potential of PolMFM for high-precision, multidimensional super-resolution imaging in complex and crowded biological environments.</p
Développement d’un système de robotique déformable pour l’imagerie in-vivo des cancers
No full textDuring Ph.D., the development of in-vivo 3D imaging based on mass spectrometry with robotic support was investigated. Indeed, mass spectrometry, which allows for distinguishing between healthy and cancerous tissues, as well as different types of tumor tissues, could open up interesting diagnostic perspectives when reconstructing this information within an image. In particular, it could improve the reliability of surgical procedures and post-operative recovery by reducing and verifying the margins around the tumor area to prevent relapses and metastases. Design aspects were addressed, including the development of a 3D imaging prototype using a rigid robot, as well as the adaptation of the Stiff-Flop robot model for mass spectrometry. Indeed, to ensure maximum safety in-vivo and to envision the automation of the procedure, deformable robotic solutions were investigated. The entire manufacturing and actuation process was studied theoretically to understand the full range of constraints associated with this robot. One of the central issues we seek to address is the precision with which we can position the robot in space for diagnostic purposes. To this end, modeling was explored and experimented in simulations to evaluate our ability to understand and predict the robot's behavior under actuation pressure. Various methods for accelerating calculations during these simulations were also investigated, to increase feedback frequency during closed-loop control. These different models were then implemented within control loops to assess the robot's performance in terms of precision and speed in various experiments. The results of these experiments were then interpreted in the context of our medical application. The integration of the tool into the operating room was also studied from the perspectives of compatibility and technical but also medical prospectsAu cours du doctorat, le développement de l’imagerie 3D in-vivo basée sur la spectrométrie de masse avec support robotique a été étudié. En effet, la spectrométrie de masse, qui permet de distinguer les tissus sains des tissus cancéreux ainsi que les différents types de tissus tumoraux, pourrait ouvrir des perspectives diagnostiques intéressantes lorsqu’on reconstruit ces informations sous forme d’image. En particulier, cela pourrait améliorer la fiabilité des interventions chirurgicales et la récupération post-opératoire en réduisant et en vérifiant les marges autour de la zone tumorale pour prévenir les rechutes et les métastases. Des aspects de conception ont été abordés, notamment le développement d’un prototype d’imagerie 3D utilisant un robot rigide ainsi que l’adaptation du modèle de robot Stiff-Flop pour la spectrométrie de masse. En effet, pour garantir une sécurité maximale en in-vivo et envisager l’automatisation de la procédure, des solutions robotiques déformables ont été étudiées. L’ensemble du processus de fabrication et d’actionnement a été étudié théoriquement afin de comprendre toutes les contraintes liées à ce robot. Une des problématiques centrales que nous cherchons à résoudre est la précision avec laquelle nous pouvons positionner le robot dans l’espace pour réaliser des diagnostics. À cette fin, différentes méthodes de modélisation ont été testées en simulation pour évaluer notre capacité à comprendre et à anticiper le comportement du robot sous pression d’actionnement. Diverses méthodes d’accélération des calculs ont également été étudiées pour augmenter la fréquence de retour d’information en boucle fermée. Ces modèles ont ensuite été intégrés dans des boucles de contrôle pour évaluer les performances du robot en termes de précision et de rapidité dans diverses expériences. Les résultats de ces expériences ont été interprétés dans le cadre de notre application médicale. Enfin, l’intégration de cet outil au bloc opératoire a été étudiée sous les angles de la compatibilité et des perspectives, tant techniques que médicales
UAV Chain Network Creation in Cluttered Environment with Flocking Rules and Routing Data
No full textInternational audienc
La défense par les chemins d’attaque: En s’appuyant sur des modèles partagés, les organisations peuvent analyser les techniques, cartographier les chemins d’attaque et bâtir une défense véritablement proactive.
No full texthttps://www.cyberun.net/la-collection#addition_content_1_10 / _11 _Cyberun#44 En s'appuyant sur des modèles partagés, les organisations peuvent analyser les techniques, cartographier les chemins d'attaque et bâtir une défense véritablement proactive.</div
Are we really ready to use radiomics and deep learning for clinical decision support in radiology?
No full textInternational audienc
“Detectors Lead, LLMs Follow”: Integrating LLMs and traditional models on implicit hate speech detection to generate faithful and plausible explanations
No full textThe journal article has been accepted and the final version is accessible free of charge from the editorial website of Elsevier at this link: https://www.sciencedirect.com/science/article/pii/S0169023X25001302International audienceSocial media platforms face a growing challenge in addressing abusive content and hate speech, particularly as traditional natural language processing methods often struggle with detecting nuanced and implicit instances. To tackle this issue, our study enhances Large Language Models (LLMs) in the detection and explanation of implicit hate speech, outperforming classical approaches. We focus on two key objectives: (1) determining whether jointly predicting and generating explanations for why a message is hateful improves LLMs' accuracy, especially for implicit cases, and (2) evaluating whether incorporating information from BERT-based models can further boost detection and explanation performance. Our method evaluates and enhances LLMs' ability to detect hate speech and explain their predictions. By combining binary classification (Hate Speech vs. Non-Hate Speech) with natural language explanations, our approach provides clearer insights into why a message is considered hateful, advancing the accuracy and interpretability of hate speech detection
Sécurité prouvable pour les noyaux et cryptographie en présence d'exécution spéculative
No full textAttacks such as Spectre and Meltdown have demonstrated that it is possible to exploit an optimization of modern processors—known as speculative execution—to compromise the security of cryptographic applications and operating system kernels. The goal of this thesis is to support the development of provably secure cryptographic applications and operating system kernels, with particular attention to the challenges posed by speculative execution. As for kernel's security, we tackle the problem of ensuring essential security guarantees such as memory safety and control flow integrity in the presence of speculative execution. To this end, we show that any kernel that is secure against attacks that do not control speculative execution can be systematically transformed into another kernel that is not subject to such attacks. We propose program transformations that enforce such a guarantee by systematically inserting speculation barriers: instructions that globally block speculative execution until their commitment. We formally demonstrate the effectiveness of these mechanisms and we evaluate their performance overhead on the Linux kernel. To improve state-of-the-art mitigations against speculative attacks, we implemented in hardware a novel speculation barrier—called dfence—that blocks speculative execution in a selective way. Our measurements show that dfence has negligible hardware overhead, and its impact on the performance of cryptographic applications is also negligible.Concerning the security of cryptographic applications, we also tackle the problem of defining a sound and complete proof system for pRHL: a relational program logic that is widely employed for proving the security of cryptographic applications. We do so, by introducing the eRHL logic: a quantitative extension of pRHL, that is complete for establishing the validity of pRHL judgments and several quantitative properties that are not directly expressible or provable within pRHL.Des attaques telles que Spectre et Meltdown ont démontré qu'il est possible d'exploiter une optimisation des processeurs modernes—connue sous le nom d'exécution spéculative—pour compromettre la sécurité des applications cryptographiques et des noyaux de systèmes d'exploitation. L'objectif de cette thèse est de favoriser le développement d'applications cryptographiques et de noyaux de systèmes d'exploitation dont la sécurité peut être démontrée formellement, avec une attention particulière portée aux défis posés par l'exécution spéculative. Concernant la sécurité des noyaux, nous abordons le problème de la garantie de propriétés de sécurité essentielles telles que la sûreté mémoire et l'intégrité du flot de contrôle en présence d'exécution spéculative. À cette fin, nous montrons que tout noyau sécurisé contre des attaques ne contrôlant pas l'exécution spéculative peut être transformé systématiquement en un autre noyau qui n'est pas vulnérable à de telles attaques. Nous proposons des transformations de programme qui assurent cette propriété en insérant systématiquement des barrières de spéculation : des instructions qui bloquent globalement l'exécution spéculative jusqu'à ce qu'elles quittent le pipeline. Nous démontrons formellement l'efficacité de ces mécanismes et évaluons leur impact sur les performances dans le noyau Linux. Pour améliorer l'état de l'art des contre-mesures contre les attaques spéculatives, nous avons implémenté une nouvelle barrière de spéculation au niveau matériel—appelée dfence—qui bloque l'exécution spéculative de manière sélective. Nos mesures montrent que dfence a un coût matériel négligeable, et son impact sur les performances des applications cryptographiques est également négligeable. En ce qui concerne la sécurité des applications cryptographiques, nous nous attaquons également au problème de la définition d'un système de preuve correct et complet pour pRHL : une logique de programme relationnelle largement utilisée pour prouver la sécurité des applications cryptographiques. Pour cela, nous introduisons eRHL, une extension quantitative de pRHL, qui est complète pour démontrer la validité des jugements de pRHL ainsi que plusieurs propriétés quantitatives qui ne sont pas directement exprimables ni démontrables dans pRHL
Fair play for individuals, foul play for groups? Auditing anonymization’s impact on ML fairness
No full textInternational audienceMachine learning (ML) algorithms are heavily based on the availability of training data, which, depending on the domain, often includes sensitive information about data providers. This raises critical privacy concerns. Anonymization techniques have emerged as a practical solution to address these issues by generalizing features or suppressing data to make it more difficult to accurately identify individuals. Although recent studies have shown that privacy-enhancing technologies can influence ML predictions across different subgroups, thus affecting fair decision-making, the specific effects of anonymization techniques, such as k-anonymity, ℓ-diversity, and t-closeness, on ML fairness remain largely unexplored. In this work, we systematically audit the impact of anonymization techniques on ML fairness, evaluating both individual and group fairness. Our quantitative study reveals that anonymization can degrade group fairness metrics by up to fourfold. Conversely, similarity-based individual fairness metrics tend to improve under stronger anonymization, largely as a result of increased input homogeneity. By analyzing varying levels of anonymization across diverse privacy settings and data distributions, this study provides critical insights into the trade-offs between privacy, fairness, and utility, offering actionable guidelines for responsible AI development. Our code is publicly available at: https://github.com/hharcolezi/anonymity-impact-fairness