Portail HAL Ensta
Not a member yet
11080 research outputs found
Sort by
From DL approach conception to operational product design : identifying roof materials for policy makers
No full textInternational audienceGiven their potential for the development of greening strategies, solar energy production or as sources of pollutants, roofs are increasingly the focus of local authorities. This study aims at producing operational roof material maps through a 12-roof material classification based on a deep learning approach applied on 5-cm orthophotos acquired over Namur, Belgium. Using convolutional neural networks and a specific data pre-processing improving the classifier learning, an overall accuracy of 81% was achieved. Operational maps built from the probabilities of belonging to each class computed by the classifier for every roof section can be used for primary estimation of specific roof material abundance and for the prioritization of areas of investigation. Areas of improvements include increasing sampling, post-classification rules, and a better management of overexposed and shaded roof sections
Advanced Topics and Smart Systems for Wireless Communications and Networks
No full textInternational audienc
Rupture et chargement cyclique
No full textInternational audienceOn présente des aspects de la mécanique de la rupture en élasticité non-linéaire.On étudie le ces de chargement cyclique pour étudier la fatigue des élastomère
Femtosecond structural probing of warm dense matter with Betatron x-ray source
No full textInternational audienceExploring and understanding ultrafast processes at the atomic level is a scientific challenge. Femtosecond X-ray Absorption Near-Edge Spectroscopy (XANES) arises as an essential experimental probing method, as it can simultaneously reveal both electronic and atomic structures, and thus potentially unravel their non-equilibrium dynamic interplay which is at the origin of most of the ultrafast mechanisms. The key point of this investigation is the achievement of a femtosecond X-ray source suitable for routine experiments. This paper will show the progressive development and improvement of such laser-plasma-based X-ray sources, starting from the picosecond down to the femtosecond scale. Time-resolved XANES measurements have been achieved and interpreted using ab initio quantum molecular dynamics simulations. This diagnostic was used to shed new light on the non-equilibrium physics involved in various materials. This paper will focus on results devoted to the non-equilibrium dynamics of a copper foil brought from solid to warm dense matter regime, by the use of a femtosecond laser pulse. Particular emphasis will be given to the recent study of the ultrafast electronic transport, which was revealed by the femtosecond time resolution
A second order asymptotic model for diffusion MRI in permeable media
No full textInternational audienceStarting from a reference partial differential equation model of the complex transverse water proton magnetization in a voxel due to diffusion-encoding magnetic field gradient pulses, one can use periodic homogenization theory to establish macroscopic models. A previous work introduced an asymptotic model that accounted for permeable interfaces in the imaging medium. In this paper we formulate a higher order asymptotic model to treat higher values of permeability. We explicitly solved this new asymptotic model to obtain a system of ordinary differential equations that can model the diffusion MRI signal and we present numerical results showing the improved accuracy of the new model in the regime of higher permeability
Double and Kubi RObots in Education: Can Telepresence Contribute to Student Engagement and Social Inclusion for Occasional Distance Learning?
No full textInternational audienc
Rebalancing gradient to improve self-supervised co-training of depth, odometry and optical flow predictions
No full textInternational audienceWe present CoopNet, an approach that improves the cooperation of co-trained networks by dynamically adapting the apportionment of gradient, to ensure equitable learning progress. It is applied to motion-aware self-supervised prediction of depth maps, by introducing a new hybrid loss, based on a distribution model of photo-metric reconstruction errors made by, on the one hand the depth + odometry paired networks, and on the other hand the optical flow network. This model essentially assumes that the pixels from moving objects (that must be discarded for training depth and odometry), correspond to those where the two reconstructions strongly disagree. We justify this model by theoretical considerations and experimental evidences. A comparative evaluation on KITTI and CityScapes datasets shows that CoopNet improves or is comparable to the state-of-the-art in depth, odometry and optical flow predictions. Our code is available here: https://github.com/mhariat/CoopNet
Probabilistic Rank and Reward: A Scalable Model for Slate Recommendation
No full textWe introduce Probabilistic Rank and Reward (PRR), a scalable probabilistic model for personalized slate recommendation. Our approach allows state-of-the-art estimation of the user interests in the ubiquitous scenario where the user interacts with at most one item from a slate of K items. We show that the probability of a slate being successful can be learned efficiently by combining the reward, whether the user successfully interacted with the slate, and the rank, the item that was selected within the slate. PRR outperforms competing approaches that use one signal or the other and is far more scalable to large action spaces. Moreover, PRR allows fast delivery of recommendations powered by maximum inner product search (MIPS), making it suitable in low latency domains such as computational advertising