Portail HAL des publications du LIRMM
Not a member yet
13279 research outputs found
Sort by
MAEVa: Un approccio ibrido per l'associazione delle variabili degli esperimenti agroecologici
No full textInternational audienceSource variables or observable properties used to describe agroecological experiments are heterogeneous, nonstandardized, and multilingual, making them challenging to understand, explain, and use in cropping system modeling and multicriteria evaluations of agroecological system performance. Data annotation via a controlled vocabulary, known as candidate variables from the Agroecological Global Information System (AEGIS), offers a solution. Text similarity measures play crucial roles in tasks such as word sense disambiguation, schema matching in databases, and data annotation. Commonly used measures include (a) string-based, (b) corpus-based, (c) knowledge-based, and (d) hybrid-based similarity, which combines two or more of these methods. This work introduces a hybrid approach called Matching Agroecological Experiment Variables (MAEVa), designed to match source and candidate variables based on (1) matching variable names, (2) matching variable descriptions, (3) a combination of (1) and (2) via a linear function, and (4) a method for selecting results for the final evaluation. For matching variable names, we propose a novel approach that extends BERT-base with an external multi-head attention layer (BERTmha). For matching variable descriptions, we augment existing descriptions using GPT-3.5 Turbo API to provide richer contextual information and employ TF-IDF to construct the vector space. Our experimental results demonstrate that BERTmha improves the precision of matching variable names by more than 11% compared to BERT-base alone, and that our constructed corpus enhances TF-IDF-based matching by more than 4%. Our evaluation (step 4) shows that MAEVa achieves a precision of over 66% from to .Les variables sources ou les propriétés observables utilisées pour décrire les expérimentations agroécologiques sont hétérogènes, non standardisées et multilingues, rendant leur compréhension, explication et utilisation difficiles dans la modélisation des systèmes de culture et les évaluations multicritères de la performance des systèmes agroécologiques. L’annotation des données via un vocabulaire contrôlé, appelé variables candidates de Agroecological Global Information System (AEGIS), constitue une solution. Les mesures de similarité textuelle jouent un rôle clé dans la désambiguïsation du sens des mots, l’appariement de schémas dans les bases de données et l’annotation des données. Les approches courantes incluent (a) la similarité fondée sur les chaînes de caractères, (b) sur le corpus, (c) sur les connaissances et (d) les approches hybrides combinant deux ou plusieurs de ces méthodes. Ce travail propose une approche hybride, Matching Agroecological Experiment Variables (MAEVa), visant à apparier les variables sources et candidates selon (1) l'appariement des noms, (2) celui des descriptions, (3) une combinaison linéaire de (1) et (2), et (4) une méthode de sélection des résultats pour l'évaluation finale. Pour l’appariement des noms, nous étendons BERT-base avec une couche d’attention multi-têtes externe (BERTmha). Pour les descriptions, nous enrichissons celles existantes avec l'API GPT-3.5 Turbo et utilisons TF-IDF pour la représentation vectorielle. Nos résultats montrent que BERTmha améliore la précision de plus de 11% par rapport à BERT-base seul et que notre corpus améliore celle de TF-IDF de plus de 4%. Notre évaluation (étape 4) montre que MAEVa atteint une précision de plus de 66% de à
Une Nouvelle Licence Professionnelle en Robotique et Intelligence Artificielle (RobIA) à l'IUT de Béziers
No full textInternational audienceDans cet article, nous décrivons, de manière brève, la nouvelle Licence Professionnelle (LP) en Robotique et Intelligence Artificielle (RobIA) qui a été conçue par l'IUT de Béziers afin de former les profils professionnels hautement qualifiés requis par l'industrie 4.0 qui est en cours de développement dans la région et le pays. Cette LP est un prototypage national, soutenu par la DGESIP, d'un potentiel nouveau BUT. La formation a accueilli la première cohorte d'étudiants en septembre 2023, elle est donc actuellement en cours de construction et d'amélioration. Dans ce contexte, l'article vise à attirer l'attention de tous les collègues enseignants qui peuvent apporter leur expérience afin d'établir correctement le programme d'études, ainsi que celle de tous les représentants des IUTs intéressés à offrir cette licence dans leurs structures, ce qui devrait potentiellement conduire à l'évolution de la formation vers un nouveau BUT
Quasi-Linear Guessing of Minimal Lexicographic Gröbner Bases of Ideals of C-Relations of Random Bi-Indexed Sequences
No full textInternational audienceComputing recurrence relations for sequences is a central problemin computer algebra, with applications in error-correcting codes, Gröbner basis computation, and sparse interpolation. While uni-indexed C-recursive sequences benefit from quasi-linear algorithms leveraging the half-gcd method, the extension to multi-indexed sequences remains computationally challenging. Existing methods for bi-indexed sequences achieve quadratic complexity at best, limiting their practical use.This paper presents a quasi-linear algorithm for computing lexicographic Gröbner bases of the ideal of C-relations associated to bi-indexed sequences. Our approach extends the half-gcd algorithm in [] by integrating a pseudo-Euclidean division. This approach shows how to leverage the bi-Hankel structure of the matrix, significantly improving the efficiency of computing minimalC-relations closing the complexity gap between the uni- and bi-indexed cases. Our algorithm is restricted to bi-indexed sequences whose associated bi-Hankel matrix has generic row rank profile
Usages d'AgroPortal dans des systèmes d'information à INRAE
No full textNational audienceAgroPortal is an ontology repository or a semantic artefact catalogue specialized in agriculture, food, and the environment. Launched in 2015, AgroPortal has become a key component of digital infrastructures to leverage semantic interoperability of data and their compliance with the FAIR principles. This article presents the uses of AgroPortal at INRAE –for its primary mission of hosting and serving semantic artefacts. Examples of integrations of these artefacts in different INRAE information systems–highlight the benefits of their use in scientific data management, annotation, & research tools.AgroPortal est un portail d'ontologies et d'artefacts sémantiques spécialisé dans les domaines de l'agriculture, de l'alimentation et de l'environnement. Lancé en 2015, AgroPortal est devenu un composant clé pour les infrastructures numériques pour favoriser l'interopérabilité sémantique des données et leur conformité aux principes FAIR. Cet article présente les usages d'AgroPortal à INRAE pour sa mission principale d'héberger et servir les artefacts sémantiques. Des exemples d'intégrations de ces ontologies dans différents systèmes d'informations à INRAE mettent en avant les bénéfices apportés par leur exploitation dans les outils de gestion, d'annotation ou de recherche de données
A genome-wide, machine learning-guided exploration of the cis-regulatory code involved in neuronal differentiation
No full textGene expression is controlled by proximal and distal cis-regulatory elements (CREs), containing DNA motifs bound by various transcription factors (TFs). Other sequence features, such as specific k-mers or low complexity regions, have also been implicated. However, in a dynamic biological process such as cell differentiation, we lack an understanding of how the transcriptional activity of CREs progressively change and what sequence features underlie these transitions, which may reflect common and/or coordinated regulatory processes. Here, we use single-cell ATAC-seq and RNA-seq to follow, at a genome scale, CREs along differentiation of induced pluripotent stem cells into cortical neurons and develop a method to automatically identify the diversity of CRE profiles and their underlying sequence features. We propose a machine-learning guided clustering algorithm, STOIC (Statistical learning TO Inform Clustering), that jointly learns an unsupervised clustering of the CREs in the space of the activity profiles and a supervised predictor associated with each cluster in the DNA-sequence space. This procedure explores the expression space and delineates the CRE clusters iteratively in order to optimize the performance of a supervised classifier predicting CRE cluster membership based on DNA sequence features. STOIC is specifically designed to provide readily interpretable results. We show that the method identifies CRE profiles associated with highly predictive sequence features and outperforms methods solely concerned with co-activity clustering on this task. Orthogonal data collected in the same settings link the inferred CRE clusters to specific enhancer or promoter signatures. Furthermore, we show that the DNA features unveiled by STOIC reflect biologically relevant regulators and offer a valuable basis to dissect elements of the cis-regulatory grammar. Finally, we demonstrate the general applicability of STOIC by analyzing five bulk CAGE datasets of human cells responding to various treatments
Local obstructions in sequences revisited
No full textIn this article we consider some simple combinatorial game and a winning strategy in this game. This game is then used to prove several known results about non-repetitive sequences and approximations with denominators from a lacunary sequence. In this way we simplify the proofs, improve the bounds and get for free the computable versions that required a separate treatment
D4.6 - Use case driven validation of semantic artefact exploitation within data repositories
No full textSemantic artifacts (SAs) such as terminologies, taxonomies, thesauri, ontologies, and metadata schemas are critical for standardising data representation and documentation, by encapsulating the meaningful knowledge within interoperability frameworks. WP4 “gathers, synthesises and disseminates the materials needed to federate the approach to metadata and ontologies at various organisational and technical levels within EOSC”. It reviews and analyses semantic artefact (SA) community practices and governance models within use cases, with the aim to explore SA management, sharing practices for data FAIRification, practices which are important for any data-driven sciences. This document reports the work of Task 4.5 on FAIR semantic artefacts in use within data repositories through nine use cases, which aim is to demonstrate the impact of FAIR SAs for (meta)data repositories. The common objective is to facilitate and encourage the use of SAs to describe and index data. To achieve this objective, discipline specific SAs catalogues are used to give access to semantically grounded, unambiguous controlled terms. The implemented solution is an API-based connector between the (meta)data repositories and the SAs catalogues (SACs) which allows data depositors to select terms from selected, relevant SAs to fill in metadata with controlled values. The use of a connector understandably benefits the data repository end users by respectively facilitating metadata filling and improving indexation quality, compared to free text values. The innovation brought by the connector profits in particular managers by preventing them from manipulating SAs (import/export, format transformation) and ensuring that their content is automatically updated. Additionally, end users gain from improved metadata quality and indexing ease. This document gathers the description of each UC and a short report of their work on connecting a SAC and a (meta)data repository. It also aims at highlighting a set of recommendations to future implementers of a connector, based on the UC experience: Identify SA and SAC capabilities of the community to meet the semantic needs in (meta)data repositories and catalogues. The organisation responsible for the SAs governance needs to be sustainable and responsive to users feedback. Connectors’ implementation induces changes in (meta)data repositories to exploit SAs potential through adapted search modalities and user interfaces. Consider users’ feedback at all stages of the connectors’ development. Take into account that the connection between SACs and (meta)data repositories impacts the ecosystem by bringing new interdependencies between information systems
Jellyfish journey live tracking using floating electronic tag
No full textInternational audienceGelatinous organisms are key players of marine ecosystems, however underlying processes of their dynamics and behaviour are still to be cleared up. Understanding the areas of production, where the blooms go and what they become are therefore of major interest in marine ecosystem management. We used floating electronic tags developed in our laboratory for jellyfish live tracking. A special attention was put on the welfare of the organisms as the tag was floating and simply attached with a fishing line around the manubrium. In situ experiments were carried out in Bages Sigean lagoon (France) where a perennial population of the Mediterranean jellyfish Rhizostoma pulmo is established. Up to 47 deployments, from 20 min to 28h, took place in 2022 and 2023 summers. Live tracking indicated that the floating device did not influence the jellyfish trajectory nor its speed. A 28-hour trajectory showed that jellyfish movement can be influenced by the wind but also by other environmental factors. The relatively small area covered by the jellyfish compared to the control float one, suggests that movements significantly influence its trajectory as a response to the environment. Jellyfish were successfully recovered suggesting in a near future repeated individual measurements processes over longer deployments
Méta-modélisation des interactions plante-plante en 3D : application à l'association colza-féverole
No full textInternational audienc
Réalité virtuelle adaptative indépendante de la tâche : Exploration et évaluation de systèmes d’aide sans connaissance de l'objectif de l’utilisateur.
No full textReal-time adaptation in virtual reality (VR) is essential for creating personalised experiences and for informing the user of the relevant actions to perform in the environment. Traditionally, when these adaptations seek to provide assistance to humans, they are task-oriented and rely on explicit knowledge of the user’s goals. In this thesis, we propose a new approach to the design of adaptive VR help systems, a task-blind design, i.e. without any knowledge of the user’s tasks or any attempt to infer them in real time. First of all, we propose a new reading of the literature by presenting a classification and modeling of adaptive systems in VR. Our classification can be broken down into three axes: the system’s knowledge level of the user’s tasks, the degree of control given to the user, and the assistance’s level provided by the systems. This enables us to put certain aspects of adaptive system design into perspective, and to propose a design model that: 1) highlights and generalises the types of input to these systems, and 2) integrates components to implement the three axes of classification. We present two studies evaluating task-blind adaptive systems that aim at enhancing cognitive abilities. The results allow us to assess 1) the feasibility of implementing this task-blind approach in complex cases, 2) the impact of such systems on user performance and behavior, and 3) the main design components that have an impact on the efficiency of such systems. This work points to promising prospects for the design of adaptive systems in VR in general, as well as more specifically for task-blind adaptive systems providing assistance.L’adaptation en temps réel en réalité virtuelle (RV) est essentielle pour créer des expériences personnalisées et pour informer l’utilisateur sur les actions pertinentes à effectuer dans l’environnement. Traditionnellement, lorsque ces adaptations cherchent à fournir une assistance à l’humain, elles sont axées sur les tâches à réaliser et reposent sur une connaissance explicite des objectifs de l’utilisateur. Dans cette thèse, nous proposons une nouvelle approche pour la conception de systèmes d’aide de RV adaptative, une conception task-blind, i.e. sans aucune connaissance des tâches de l’utilisateur ou tentative de les déduire en temps réel. Tout d’abord, nous proposons une nouvelle lecture de la littérature en présentant une classification et une modélisation des systèmes adaptatifs en RV. Notre classification se décline en trois axes : le degré de connaissance du système envers les tâches de l’utilisateur, le degré de contrôle donné à l’utilisateur et le niveau d’aide fourni par les systèmes. Cela nous permet de mettre en perspective certains aspects de la conception de systèmes adaptatifs et de proposer un modèle de conception qui : 1) met en lumière et généralise les types d’entrées de ces systèmes, et 2) intègre des composants permettant de mettre en oeuvre les trois axes de la classification. Nous présentons deux études évaluant des systèmes adaptatifs task-blind qui visent à améliorer les capacités cognitives. Les résultats permettent d’évaluer 1) la faisabilité de l’implémentation de notre approche task-blind dans des cas complexes, 2) l’impact de tels systèmes sur la performance et le comportement des utilisateurs, et 3) les composants de conception principaux qui ont un impact sur l’efficacité de tels systèmes. Ces travaux laissent envisager des perspectives prometteuses sur la conception des systèmes adaptatifs en RV en général, ainsi que plus spécifiquement sur les systèmes adaptatifs task-blind fournissant une assistance