Archive ouverte de Centrale Lyon
Not a member yet
32420 research outputs found
Sort by
Large Language Models and Algorithm Execution: Application to an Arithmetic Function
No full textLarge Language Models (LLMs) have recently developed new advanced functionalities. Their effectiveness relies on statistical learning and generalization capabilities. However, they face limitations in internalizing the data they process and struggle, for instance, to autonomously execute algorithms. In this paper, we investigate the possibility of extending these models' capabilities to algorithm execution through specialized supervised training focused on reasoning decomposition. We introduce a training model called LLM-DAL (Large Language Model - Decompositional Algorithmic Learning), through which we demonstrate that LLMs' ability to perform complex algorithmic inferences and generalize can be significantly improved when the training method is properly designed to guide the model in its learning process
Shear mechanical properties measurements at the surface scale: Enhanced performances of the micro-shear compression specimen
No full textInternational audienceAn intensive study combining experimental tests and numerical simulations was carried out to improve the understanding of the micro-shear test using the Micro-shear Compression Specimen (MCS). The results demonstrated good data reliability in the elastic regime up to the yield stress. However, the study also revealed that friction between the flat punch and the MCS significantly affects the plastic regime, and must therefore be accounted for to accurately extract shear mechanical properties. To overcome this limitation, two alternative methods were developed. The first one consists in compressing a new type of micro-shear compression specimen, featuring two perpendicular gauges forming a cross geometry (X-MCS). The second consists of applying multicycle loading to the conventional MCS. Both approaches successfully eliminated friction dependence in the plastic regime, in contrast to the classical method. Finally, the X-MCS geometry was applied to very high strain rate testing on fused silica. Thanks to the small gauge height of the X-MCS, it was possible to measure shear mechanical properties at a strain rate of 104 s−1, which was not achieved using conventional micropillar compression with our micromechanical setup. These methods provide a new pathway for extracting shear mechanical properties, which are critical in the field of tribology, where surfaces are subjected to intense shear deformation
QuaQue : conception et implémentation SQL d'une algèbre condensée pour la gestion concurrente des versions des graphes de connaissances
No full textInternational audienceThe management of versioned knowledge graphs presents significant challenges, particularly in querying data across multiple versions efficiently. This paper introduces QuaQue, a key component of the ConVer-G system, which addresses this challenge by translating SPARQL (SPARQL Protocol and RDF Query Language) queries into SQL (Structured Query Language). QuaQue leverages a novel condensed algebra to operate on a relational model where versioning information is compactly stored using bitstrings. This approach allows for efficient querying of concurrent versions of knowledge graphs within a standard relational database system. We present the key concepts of our condensed algebra, detail the translation process from SPARQL algebra to SQL, and provide a comparative benchmark against a native RDF (Resource Description Framework) triple store, demonstrating the viability and performance benefits of our approach.La gestion des graphes de connaissances versionnés pose des défis importants, notamment en ce qui concerne l'interrogation efficace des données sur plusieurs versions. Cet article présente QuaQue, un composant clé du système ConVer-G, qui relève ce défi en traduisant les requêtes SPARQL (SPARQL Protocol and RDF Query Language) en SQL (Structured Query Language). QuaQue exploite une nouvelle algèbre condensée pour opérer sur un modèle relationnel où les informations de versionnement sont stockées de manière compacte à l'aide de chaînes de bits. Cette approche permet d'interroger efficacement des versions concurrentes de graphes de connaissances au sein d'un système de base de données relationnelle standard. Nous présentons les concepts clés de notre algèbre condensée, détaillons le processus de traduction de l'algèbre SPARQL vers le SQL, et fournissons un benchmark comparatif par rapport à un triplet store RDF (Resource Description Framework) natif, démontrant la viabilité et les avantages en termes de performances de notre approche
Open-vocabulary models for object detection and segmentation in visual art: survey and comparative study
No full textInternational audienceObjects present in paintings help art history specialists interpret and decode artworks. The analysis of large, digitized artistic collections became feasible thanks to modern object detection approaches. Nevertheless, the use of object detection models typically requires fine-tuning for specific tasks. Therefore, art history specialists are remain constrained by the categories of objects in existing labeled artistic datasets when using artificial intelligence methods. This limitation can be overcome by using recent models that combine two modalities: vision and text. Vision-language models have made open-vocabulary detection (OVD) possible, allowing detection without restrictions on the applied categories, in contrast to fixed-vocabulary detection. Recent literature lacks a comprehensive review focusing on OVD in artistic images. In this paper we analyze state-of-the-art models for OVD, analyze their transferability to cultural heritage categories and systematically evaluate them on artistic datasets commonly used in literature. The DEArt and IconArt datasets, which are annotated with cultural heritage-specific categories contain paintings from the 11th to the 20th century. While the Watercolor2K dataset, annotated with common object categories consists of watercolor paintings. Based on our analysis, the OWLv2 model achieved the best performance in both object detection and grounding task scenarios on these datasets. Additionally, we discuss existing challenges of open-vocabulary segmentation in artistic images and future tasks
On the partial Grundy coloring of graphs
No full textInternational audienceIn a proper vertex coloring of a graph, a Grundy vertex is a vertex colored with color c and adjacent to any color 1 to c-1. A partial Grundy coloring of a graph is a proper coloring of its vertices where any color admits at least one Grundy vertex. The partial Grundy number of a graph is the maximum number of colors used in a partial Grundy coloring of this graph. In this article we consider this parameter for some classes of graphs, in particular -free graphs, regular graphs, Cartesian and direct products of graphs
Quasi-geostrophic Rayleigh-Bénard convection on the tilted -plane
No full textInternational audienceRapidly rotating Rayleigh-Bénard convection on a -plane at colatitude is investigated numerically using an asymptotically reduced equation set valid in the limit of very rapid rotation. The equations provide a non-hydrostatic but quasi-geostrophic description in a non-orthogonal coordinate system. The tilt changes the structure of the large-scale barotropic condensate from large-scale vortices to zonal flows as the colatitude of the -plane increases, with bistable states present for certain parameter ranges, extending prior work to a geophysically significant parameter regime. This behaviour is understood through the impact of broken rotation symmetry on the barotropic source terms resulting from baroclinic vortical stresses and baroclinic torque. As the tilt angle increases, global heat and momentum transport is reduced relative to upright-polar convection, a result that is explained through linear theory and nonlinear power maps both of which demonstrate increased attenuation of the domain of dynamically active spatial scales as the convective modes depart from a North-South alignment in the horizontal plane. A key finding is that the predominance of lateral thermal mixing allows for the maintenance of a persistent unstable mean temperature gradient that saturates at increasing forcing levels and remains insensitive to the colatitude
On some Fraïssé limits with free amalgamation
No full textIn this work a general way is given to construct some examples of NSOP1 theories aslimits of some Fraïssé class satisfying strong conditions. These limits will satisfy existence, that Kimindependence coincides with algebraic independence, and that forking independence is obtained byforcing base monotonicity on Kim-forking. These theories also come with a stationary independencerelation. This study is based on results of Baudisch, Chernikov, Kruckman and Ramsey
Factors Associated With Progression, Resolution and Mortality of Patients With Overt Hepatic Encephalopathy
No full textInternational audienceOvert hepatic encephalopathy (OHE) is a reversible complication of cirrhosis that often results in hospitalization. Factors associated with progression, resolution and mortality are not known, particularly with confounders such as acute-on-chronic liver failure (ACLF). The aim of the study was to evaluate factors associated with progression, resolution, and mortality of patients with OHE. Methods: Data for this study were derived from PREDICT, a prospective cohort study of patients with cirrhosis hospitalized for an acute decompensation or ACLF. Progression to OHE or worsening in severity and resolution from OHE were evaluated at 1 week. Cox regression, interaction analyses, and Kaplan-Meier curves were performed. Results: One thousand two hundred seventy-three patients were included [68% males; 59 (51-67) years; 56% alcohol], 16% admitted with OHE and 16% with ACLF. Older age, metabolic dysfunction-associated steatotic liver disease, previous treatment with lactulose, ACLF, white blood cell counts or albumin levels at admission were associated with OHE (P < 0.05). OHE progressed in 3% patients, which was associated with older age, previous treatment with lactulose and bacterial infections (P < 0.05), with a significantly shorter time-to-death (P < 0.001). Patients who resolved OHE (79%) presented a similar prognosis than those without OHE (P = 0.208). Post hoc analysis of the age-adjusted interaction between OHE and ACLF to predict mortality showed higher differences across ACLF grades compared with OHE. Conclusion: Presence of ACLF and progression of OHE are associated with high short-term mortality rates, while resolution of OHE is associated with significantly better prognosis. Understanding the natural history of OHE will have profound implications on the development of novel approaches
Real-Time Simulation of Deformable Tactile Sensors and Objects in Robotic Grasping using Graph Neural Networks with Inductive Biases
No full textInternational audiencePhysical simulation of deformable bodies is crucial for robotic manipulation, particularly for applications involving deformable objects and deformable tactile sensors. While Finite Element Method (FEM) simulations provide high accuracy for modeling deformable objects and tactile sensors, they are prohibitively expensive for real-time applications, with simulation times often exceeding practical limits for robotic control and learning. This paper presents a novel Graph Neural Network (GNN) framework that accelerates the simulation of tactile sensors by factors of - compared to FEM, while maintaining high physical accuracy. Our approach addresses limitations in existing GNN-based physics learning through inductive biases. The key contributions include: (1) extending FEM simulation to deformable tactile sensors in grasping scenarios, (2) incorporating novel inductive biases through tetrahedral features and global graph features to improve information propagation and training stability, and (3) demonstrating the first successful application of GNN simulation for tactile sensors with generalization to unseen objects. Additionally, the inductive biases reduce prediction errors by up to 45% compared to baseline approaches. This work enables real-time soft tactile sensors of soft object simulation for robotic applications with stress prediction. All simulation and training code will be released