Portail HAL Ensta
Not a member yet
11080 research outputs found
Sort by
SpatialSim: Recognizing Spatial Configurations of Objects with Graph Neural Networks
No full textRecognizing precise geometrical configurations of groups of objects is a key capability of human spatial cognition, yet little studied in the deep learning literature so far. In particular, a fundamental problem is how a machine can learn and compare classes of geometric spatial configurations that are invariant to the point of view of an external observer. In this paper we make two key contributions. First, we propose SpatialSim (Spatial Similarity), a novel geometrical reasoning benchmark, and argue that progress on this benchmark would pave the way towards a general solution to address this challenge in the real world. This benchmark is composed of two tasks: Identification and Comparison, each one instantiated in increasing levels of difficulty. Secondly, we study how relational inductive biases exhibited by fully-connected message-passing Graph Neural Networks (MPGNNs) are useful to solve those tasks, and show their advantages over less relational baselines such as Deep Sets and unstructured models such as Multi-Layer Perceptrons. Finally, we highlight the current limits of GNNs in these tasks
Invariant integrals and asymptotic fields near the front of a curved planar crack
No full textInternational audienceA plane crack is considered and the influence of local curvature of the crack front on the local mechanical fields is studied. The main goal is to determine the stress intensity factors along a curved planar crack in linear elasticity with accuracy. This is obtained by determination of new test fields and the use of bilinear forms, issued from invariant integrals, which separate the local modes of fracture
Multi-Agent Reinforcement Learning as a Computational Tool for Language Evolution Research: Historical Context and Future Challenges
No full textInternational audienceComputational models of emergent communication in agent populations are currently gaining interest in the machine learning community due to recent advances in Multi-Agent Reinforcement Learning (MARL). Current contributions are however still relatively disconnected from the earlier theoretical and computational literature aiming at understanding how language might have emerged from a prelinguistic substance. The goal of this paper is to position recent MARL contributions within the historical context of language evolution research, as well as to extract from this theoretical and computational background a few challenges for future research
Martingale driven BSDEs, PDEs and other related deterministic problems
No full textInternational audienceWe focus on a class of BSDEs driven by a cadlag martingale and corresponding Markov type BSDE which arise when the randomness of the driver appears through a Markov process. To those BSDEs we associate a deterministic problem which, when the Markov process is a Brownian diffusion, is nothing else but a parabolic type PDE. The solution of the deterministic problem is intended as decoupled mild solution, and it is formulated with the help of a time-inhomogeneous semigroup
Automatic Curriculum Learning For Deep RL: A Short Survey
No full textInternational audienceAutomatic Curriculum Learning (ACL) has become a cornerstone of recent successes in Deep Reinforcement Learning (DRL). These methods shape the learning trajectories of agents by challenging them with tasks adapted to their capacities. In recent years, they have been used to improve sample efficiency and asymptotic performance, to organize exploration, to encourage generalization or to solve sparse reward problems, among others. To do so, ACL mechanisms can act on many aspects of learning problems. They can optimize domain randomization for Sim2Real transfer, organize task presentations in multi-task robotic settings, order sequences of opponents in multi-agent scenarios, etc. The ambition of this work is dual: 1) to present a compact and accessible introduction to the Automatic Curriculum Learning literature and 2) to draw a bigger picture of the current state of the art in ACL to encourage the cross-breeding of existing concepts and the emergence of new ideas
Modeling A UAV in Practice: A Comparison between Rhapsody and Capella
No full textInternational audienceMBSE (Model-Based System Engineering) is the formalisedapplication of modeling to support system requirements,design, analysis, verification and validation activities beginningin the conceptual design phase and continuing throughout developmentand later life cycle phases. This paper is a practicalapproach of MBSE methods and tools comparison. It introducesa new COmparative FRAmework, named COFRA, that aims atputting MBSE methods and tools to test. The article proposes theevaluation of the ARCADIA method and its tool Capella and theevaluation of Nexter’s custom-made PROXIMITY method usingRhapsody tool. We have detailed the motivations for choosingthese tools and methods as a stepping stone for MBSE toolsbenchmark. The article presents a few practical representationsof an electrical UAV, which is an example of multi-physical systemthat suits our interrogations about methods and tool comparisons.Then, this paper concludes with our future works, with the needto anchor systems engineering with formal comparative analysisfor MBSE, and how to tailor the methods and tools to thesystem purpose and category. This introduces the requirement ofsystem-driven engineering, so that development costs and delaysbe formally related to business income
A hyperbolic phase-transition model coupled to tabulated EoS for two-phase flows in fast depressurizations
No full textInternational audienc
Liquefaction triggering in silty sands: effects of non-plastic fines and mixture-packing conditions
No full textInternational audienceDuring recent seismic events, such as 2010 Darfield and 2016 Ecuador earthquakes, widespread liquefaction has been observed in sand deposits with silt content. Nevertheless, the presence of non-plastic fines implies variable liquefaction resistance of sands.The goal of this research is the assessment of the influence of non-plastic fines and mixture-packing conditions on liquefaction triggering. A series of monotonic and cyclic consolidated undrained triaxial as well as resonant column tests is carried out on reconstituted soil specimens. The behavior of loose, medium and dense silty-sands is analyzed, using different fine contents and confining pressures. The results show that the behavior of mixtures strongly depends on the packing configuration of coarse and fine particles. The experimental results are analyzed in terms of equivalent intergranular void ratio, which is identified in the literature as an adequate state parameter to characterize the global effect of fine particles. The estimation of the equivalent intergranular void ratio requires the determination of the active fine fraction participating in the force transfer. An original formula is proposed for the parameter based on packing configuration. The validation of the proposed formula is undertaken through comparisons with the present experimental results but also with results reported in the literature. The proposed expression to estimate the active fine fraction allows a satisfactory prediction of liquefaction triggering in sand-fines mixtures independently from the fine content
