Portail HAL U-Bordeaux-Montaigne
Not a member yet
46786 research outputs found
Sort by
Méthode Melville, conception d’un outil convivial pour la recherche-création
No full textInternational audienc
Abu Bara, Leader of Ansaru: The Life and Capture of a Veteran Nigerian Jihadi Militan
No full textArticle published on the Section « Community Insights » of SARI Global websiteThis article examines the life and career of the recently captured leader of Ansaru, Mahmud Muhammad Usman, alias Abu Bara, drawing from a set of interviews and other related sources the authors have been compiling for a wider project on the origins of the “Boko Haram” phenomenon. It provides readers with contextual information to better understand the significance of this arrest and its implications
Saisir le panafricanisme depuis la « marge marocaine ». Une entrée par les conférences des peuples africains (1958-1961)
No full textInternational audienc
Châtelperronian cultural diversity at its western limits: Shell beads and pigments from La Roche-à-Pierrot, Saint-Césaire
No full textInternational audienceThe timing and mechanisms underlying the gradual replacement of Neanderthal populations by Homo sapiens groups have elicited heated debate for decades. The Châtelperronian of France and northern Spain has figured prominently in these discussions. While the Neanderthal authorship of this techno-complex is far less certain than previously thought, the Châtelperronian is now generally accepted as being among the first genuine Upper Paleolithic industries in Eurasia rather than a \"transitional\" stone tool technology. Here, we report the first association of shell beads with Châtelperronian artifacts from new excavations at the site of La-Roche-à-Pierrot, Saint-Césaire (France), supported by a multiproxy spatial analysis that incorporates site formation processes. Similar types of personal ornaments are unknown from Middle Paleolithic contexts. Comparable examples have, however, been recovered from contexts penecontemporaneous with the Châtelperronian in south-eastern Europe and around the Mediterranean. This hitherto undocumented combination of an early Upper Paleolithic industry and shell beads provides insights into cultural variability in western Europe and raises the question as to whether the makers of the Châtelperronian were influenced by or formed part of the earliest dispersals of H. sapiens into the region
ZOTERORAGpy : Système de veille scientifique et de gestion de connaissances pour la recherche académique
No full textRAGpy constitutes a software environment dedicated to the automation and optimization of scientific monitoring practices within the context of academic research. Developed in response to the increasing documentary management constraints faced by researchers and doctoral students, this tool provides an integrated processing pipeline enabling the transformation of heterogeneous bibliographic corpora into structured and queryable knowledge bases.The RAGpy architecture is organized around five distinct functional components:1. The first concerns document ingestion, handling data import from bibliographic managers (Zotero) or tabular files, coupled with textual extraction through optical character recognition (Mistral OCR).2. The second component performs semantic segmentation of documents into calibrated textual segments, following established standards in the field of natural language processing.3. The third phase generates dense vector representations (OpenAI embeddings, 3072 dimensions) and sparse representations (TF-IDF via spaCy), enabling the capture of semantic properties of content.4. These representations subsequently feed two exploitation modalities: on one hand, indexing in vector databases (Pinecone, Weaviate, Qdrant) intended for implementing Retrieval-Augmented Generation (RAG) systems; on the other hand, automatic classification through unsupervised clustering (UMAP + HDBSCAN) enabling the identification of thematic groupings within the corpus.5. A fifth functionality ensures the automated generation of structured reading notes, synchronized with the user's Zotero library.From a technical infrastructure perspective, RAGpy is built on the FastAPI framework and integrates a JWT token-based authentication system. The application can be deployed via Docker containerization, facilitating its installation and reproducibility across different working environments.The targeted use cases primarily concern the conduct of systematic literature reviews, thematic analysis of large documentary corpora, and the constitution of augmented memories enabling the querying of scientific collections through language models. RAGpy thus situates itself within the field of digital humanities and computational methods applied to the social sciences and humanities, proposing a technical apparatus intended to support scholarly reading practices and bibliographic organization.RAGpy constitue un environnement logiciel dédié à l'automatisation et à l'optimisation des pratiques de veille scientifique dans le contexte de la recherche académique. Développé en réponse aux contraintes croissantes de gestion documentaire auxquelles sont confrontés chercheurs et doctorants, cet outil propose une chaîne de traitement intégrée permettant la transformation de corpus bibliographiques hétérogènes en bases de connaissances structurées et interrogeables.L'architecture de RAGpy s'articule autour de cinq composantes fonctionnelles distinctes : 1. La première concerne l'ingestion documentaire, qui prend en charge l'importation de données issues de gestionnaires bibliographiques (Zotero) ou de fichiers tabulaires, couplée à une extraction textuelle par reconnaissance optique de caractères (OCR Mistral). 2. La deuxième composante assure le découpage sémantique des documents en segments textuels calibrés, suivant les standards établis dans le domaine du traitement automatique des langues. 3. La troisième phase génère des représentations vectorielles denses (embeddings OpenAI, 3072 dimensions) et creuses (TF-IDF via spaCy), permettant la capture des propriétés sémantiques des contenus. 4. Ces représentations alimentent ensuite deux modalités d'exploitation : d'une part, l'indexation dans des bases vectorielles (Pinecone, Weaviate, Qdrant) destinées à l'implémentation de systèmes de génération augmentée par récupération (RAG) ; d'autre part, une classification automatique par clustering non supervisé (UMAP + HDBSCAN) permettant l'identification de regroupements thématiques au sein du corpus. 5. Une cinquième fonctionnalité assure la génération automatisée de fiches de lecture structurées, synchronisées avec la bibliothèque Zotero de l'utilisateur.Du point de vue de l'infrastructure technique, RAGpy repose sur le framework FastAPI et intègre un système d'authentification par jetons JWT. L'application peut être déployée via conteneurisation Docker, facilitant son installation et sa reproductibilité dans différents environnements de travail.Les cas d'usage visés concernent principalement la conduite de revues de littérature systématiques, l'analyse thématique de corpus documentaires volumineux, et la constitution de mémoires augmentées permettant l'interrogation de collections scientifiques via des modèles de langage. RAGpy s'inscrit ainsi dans le champ des humanités numériques et des méthodes computationnelles appliquées aux sciences humaines et sociales, en proposant un dispositif technique destiné à accompagner les pratiques de lecture savante et d'organisation bibliographique
Parcours de la bande dessinée de science-fiction aux États-Unis
No full textInternational audienceLa bande dessinée de science-fiction américaine a produit des images inoubliables, des paysages de Mongo dans Flash Gordon à l’arrivée de Galactus à New York dans Les Quatre Fantastiques. Ce bref panorama vise à présenter les grands mouvements et les grandes œuvres qui ont façonné le genre
A Fine-Grained Predictive Stress Quantification Framework for Drivers with Autism Spectrum Disorder in Inclusive Smart Mobility
No full textInternational audienceAutism Spectrum Disorder (ASD) is a neurodevelopmental condition that significantly heightens the challenges and risks associated with driving. Individuals with ASD often experience increased sensory sensitivities, difficulty adapting to dynamic environments, and heightened vulnerability to unexpected events, which can lead to elevated stress levels, sensory overload, and impaired decision-making on the road. Existing research in stress detection primarily proposes reactive approaches, triggering responses only after stress has escalated. Such strategies limit the possibility for timely and effective intervention. Moreover, most existing studies fail to quantify stress intensity, providing only binary assessments. This lack of granularity hinders the development of proactive and personalized support mechanisms. In this paper, we propose a novel real-time approach for continuous stress prediction and quantification tailored to drivers with ASD. Our method leverages physiological signals and contextual data to anticipate stress escalation and enable proactive stress quantification and provide a proactive, fine-grained assessment of stress levels, paving the way for personalized and context-aware in-vehicle support systems that enhance driving safety and comfort for individuals with ASD