Portail HAL des publications du LIRMM
Not a member yet
13279 research outputs found
Sort by
Analyse de l'encodeur pour la segmentation d'une base de données hétérogène de photographies de plaies chroniques avec peu d'annotations
No full textInternational audienceSegmentation task is crucial in medical imaging, but obtaining a sufficient quantity of annotated data is challenging, limiting the development of high-performing deep learning models. Self-supervised learning (SSL) strategies offer a promising solution to address this lack of annotation. One such strategy, Dinov2 for Distillation with NO labels, enabled the creation of the vast LVD-142M database and als the training of encoders, whose weights are now freely accessible. However, clinical images may not be well represented in LVD-142M. In this article, we evaluate the benefits of different encoder initialization methods for segmentation in a context of scarce annotated clinical data.La segmentation est cruciale en imagerie médicale mais l’obtention de données annotées en quantité suffisante est difficile, limitant le développement de modèles d’apprentissage profond performants. Les stratégies d’apprentissage auto-supervisé (SSL) offrent une solution prometteuse pour pallier ce manque d’annotation. L’une d’entre elles, Dinov2 pour Distillation with NO labels a permis l’élaboration de l’immense base de données LVD-142M ainsi que l’entraînement d’encodeurs, aujourd’hui en accès libre. Cependant, les images cliniques ne sont pas nécessairement bien représentées dans LVD-142M. Dans cet article, nous comparons différentes méthodes d’initialisation d’encodeurs pour la segmentation de photographies cliniques dans un contexte de manque d’annotation
FAIR-IMPACT project response to "DRAFT: Developing and implementing the semantic interoperability recommendations of the EOSC Interoperability Framework"
No full textFAIR-IMPACTIn January 2024, the EOSC Association’s Task Force on Semantic Interoperability published a draft report called “Developing and implementing the semantic interoperability recommendations of the EOSC Interoperability Framework”. This report aims to enrich and refine some concepts covered by the semantic interoperability aspects within the EOSC Interoperability Framework (EOSC-IF) published in 2021. FAIR-IMPACT aims to enable FAIR by supporting the uptake of tools, approaches and methods that are coming from various sources. This support is organised across scientific communities and at multiple levels (institutional, national and European). As the FAIR principles and cross-disciplinary efforts heavily rely on semantic interoperability, FAIR-IMPACT is keen to discuss the expansion and wording of related concepts through this open EOSC community discussion. This report describes the views and thoughts coming from FAIR-IMPACT work packages (WP) members dealing with semantic artefacts and interoperability, which correspond respectively to Metadata and Ontologies (WP4) and to Interoperability (WP6). WP4 is actively involved in scrutinising the use of semantic artefacts from different aspects: their governance, their design, their catalogues, their use for research software, their mappings and their uses along with data repositories. WP6 is striving to design and promote interoperability mechanisms across domains and institutions and foster global alignment of FAIR frameworks. By working in concert both WPs lead to better semantic interoperability with and between disciplines, and foster semantic artefact harmonisation across the EOSC ecosystem. The FAIR-IMPACT response is organised in five main sections according to the five stated recommendations by the EOSC association Task Force on Semantic Interoperability report, which are: Align emerging adaptations and implementations to the semantic view of the EOSC-IF reference architecture. Identify and consolidate different approaches to representing and exchanging (meta)data with the FDO model described in the EOSC-IF. Extend the EOSC-IF to include a research process perspective that can support convergence on solutions for common use cases. Extend the set of semantic objects to include artefacts such as mappings and crosswalks. Recognise the semantic artefact catalogue as a critical part of the long-term viability of any research data infrastructure. Our methodology was: WP4 and WP6 members have been invited to express their thoughts on the suggested recommendations according to their expertise and experience in FAIR-IMPACT. Thus, WP6 members mostly reviewed the interoperability aspects, specifically the first three recommendations while WP4 members mostly focused their efforts on the last two recommendations in semantic objects. But this is not exclusive. From the WP members individual points of view, feedback was gathered and synthesised, recommendation by recommendation, by one author. The overall document was then reviewed again by WP members and the FAIR-IMPACT executive team
Communication Complexity of the Secret Key Agreement in Algorithmic Information Theory
No full textPreliminary version was published in proceedings MFCS 2020, 14 pages.http://mfcs.mff.cuni.cz/2020/ https://drops.dagstuhl.de/entities/document/10.4230/LIPIcs.MFCS.2020.44 Journal version: https://dl.acm.org/doi/10.1145/3665163 (36 pages, 8 figures)arxIv version: arXiv:2004.13411 (32 pages, 8 figures)International audienceIt is known that the mutual information, in the sense of Kolmogorov complexity, of any pair of strings x and y is equal to the length of the longest shared secret key that two parties can establish via a probabilistic protocol with interaction on a public channel, assuming that the parties hold as their inputs x and y respectively. We determine the worst-case communication complexity of this problem for the setting where the parities can use private sources of random bits. We show that for some pairs x, y the communication complexity of the secret key agreement does not decrease even if the parties have to agree on a secret key whose size is much smaller than the mutual information between x and y. On the other hand, we discuss examples of x, y such that the communication complexity of the protocol declines gradually with the size of the resulting secret key. The proof of the main result uses spectral properties of appropriate graphs and the expander mixing lemma, as well as information theoretic inequalities
Temporalizing Digraphs via Linear-Size Balanced Bi-Trees
No full textInternational audienceIn a directed graph D on vertex set v₁,… ,v_n, a forward arc is an arc v_iv_j where i 0 such that one can always find an enumeration realizing c.|R| forward connected pairs {x_i,y_i} (in either direction)
Planning Impact-Driven Logistic Tasks
No full textInternational audienceThis letter proposes a decoupled two-level modelbased planning strategy and control for a class of robotic tasks involving impact to be made with desired performance and constraints. The first part solves the problem of planning a robotic arm trajectory from a given state (position and velocity) to another desired one, enabling non-stop trajectory cycles. The second part is an impact-aware model-based plugin; it is specific to each task and links the desired task-space impact objective to a via-point in the joint-space. The two parts are then combined to achieve the entire task. Our approach is assessed with realrobot experiments demonstrating how this strategy can be used to perform tossing, grabbing, and boxing or any combination of them in sorting logistics-industry use-cases
Reducing the Silicon Area Overhead of Counter-Based Rowhammer Mitigations
No full textInternational audienceModern computer memories have shown to have reliability issues. The main memory is the target of a security threat called Rowhammer, which causes bit flips in adjacent victim cells of aggressor rows. Numerous countermeasures have been proposed, some of the most efficient ones relying on row access counters, with different techniques to reduce the impact on performance, energy consumption and silicon area. In these proposals, the number of counters is calculated using the maximum number of row activations that can be issued to the protected bank. As reducing the number of counters results in lower silicon area and energy overheads, this can have a direct impact on the production and usage costs. In this work, we demonstrate that two of the most efficient countermeasures can have their silicon area overhead reduced by approximately 50% without impacting the protection level by changing their counting granularity
Integrating data and knowledge to support the selection of service plant species in agroecology
No full textInternational audienceHighlights: • Plant functional traits can be linked to ecosystem services (ES) using a logical rule-based language. • An ontology-based data access approach enables to identify the most relevant species for a desired ES. • The management of missing and redundant trait values in databases is key to the reliability of species selection. • The proposed tool ranks service plant species similarly to published and expert knowledge.Abstract: There is a crucial need for tools to help researchers, technicians and farmers designing sustainable agroecosystems based on agroecology Indeed, such agroecosystems are inherently complex and their design requires to integrate various data and unstabilised scientific knowledge. In this paper, we consider the issue of selecting service plant species according to their potential to provide ecosystem services. To tackle that issue, we adopt an approach based both on a formalized representation of domain knowledge, which enables reasoning, and on the exploitation of available data, collected independently of the targeted application. More specifically, we rely on the one hand on recent scientific results in agronomy linking functional traits (i.e., measurable characteristics of plant species) to ecosystem services, and on the other hand on data about functional traits collected by the research community in ecology. The architecture of our system is inspired by the ontologybased data access paradigm, which allows to combine data and knowledge in a principled way. We provide a methodology to acquire scientific knowledge in the form of diagrams linked to data sources, as well as a formalization in a logical rule-based language. Importantly, our rules are independent from specific diagrams and data, to ensure genericity and facilitate the evolution of the system. We detail the construction of a knowledge base devoted to vine grassing, i.e., installing herbaceous service plants in vineyards, and present an evaluation of the system's results on this use case. We finally discuss the lessons learned and further challenges to be met
A maturity model for catalogues of semantic artefacts
No full textInternational audienceThe work presented in this paper is twofold. On the one hand, we aim to define the concept of semantic artefact catalogue (SAC) by overviewing various definitions used to clarify the meaning of our target of observation, including the meaning of the focal item: semantic artefacts. On the other hand, we aim to identify metrics and dimensions that can be used to assess the maturity of such catalogues. In particular, we define a maturity model to measure, compare and evaluate available semantic artefact catalogues. By setting these dimensions and their metrics, catalogues can be classified by each dimension. So the maturity of both the catalogues and the dimensions as a whole can be expressed. Such a maturity model and its application to 26 semantic artefacts catalogues - from various disciplines and relying on various technologies - are available to be later enriched
Non-Volatile Latch Designs with Node-Upset Tolerance and Recovery using Magnetic Tunnel Junctions and CMOS
No full textInternational audienceAs semiconductor technologies scale down, radiativeparticle-induced soft errors and static power consumption are becoming major concerns for digital circuits. Magnetic-tunneljunctions (MTJs) are widely used to address these concerns. MTJs are non-volatile and compatible with traditional CMOS processes. In this paper, we first propose a double-node-upset (DNU) tolerant and non-volatile latch, i.e., M-TPDICE-V2, providing high reliability. Additionally, we further propose an advanced latch, namely M-8C, that is able to completely recover from single-nodeupsets (SNUs) and DNUs. M-8C uses a DNU recovery module and a backup and restore module based on a pair of MTJs. Furthermore, we propose a universal backup and restore module suitable for any latch providing non-volatility. We simulate the proposed latches using the Synopsys HSPICE tool with a 45nm CMOS process model. Simulation results confirm the superior capabilities of our proposed M-TPDICE-V2 and M-8C latches. MTPDICE-V2 exhibits strong SNU and DNU tolerance and nonvolatility, while the M-8C latch provides complete DNU recovery capabilities
Le contexte en traitement automatique des langues
No full textInternational audienceCet article essaie de définir ce qu'est un contexte en traitement automatique des langues, comment se calcule le contexte et comment le contexte aide au calcul des analyses. La notion même de contexte a été profondément modifiée par les approches récentes, quantitatives, du traitement automatique du langage naturel lesquelles sont principalement motivées par la réalisation d'applications efficaces. C'est pourquoi notre article se divise en deux partie, l'une consacrée à la notion de contexte dans le TAL symbolique, qui distingue différentes notions de contexte correspondant aux différents niveaux d'analyse linguistique ; l'autre, fondée sur des algorithmes quantitatifs (des méthodes probabilistes, puis d'apprentissage automatique) conduit à des notions numériques de contexte dont le lien avec tel ou tel niveau d'analyse linguistique, moins immédiat, nécessite un travail d'interprétation. Après ces deux parties, nous conclurons par une section sur le contexte dans le TAL hybride qui voit un système de traitement en une succession d'étapes de machine learning conduisant chacune à une structure numérique intermédiaire interprétable