HAL Portal UBS (Université Bretagne Sud)
Not a member yet
14201 research outputs found
Sort by
A Generative Artificial Intelligence framework for long-time plasma turbulence simulations
No full textInternational audienceGenerative deep learning techniques are employed in a novel framework for the construction of surrogate models capturing the spatio-temporal dynamics of 2D plasma turbulence. The proposed Generative Artificial Intelligence Turbulence (GAIT) framework enables the acceleration of turbulence simulations for long-time transport studies. GAIT leverages a convolutional variational auto-encoder and a recurrent neural network to generate new turbulence data from existing simulations, extending the time horizon of transport studies with minimal computational cost. The application of the GAIT framework to plasma turbulence using the Hasegawa-Wakatani (HW) model is presented, evaluating its performance via various analyses. Very good agreement is found between the GAIT and the HW models in the spatio-temporal Fourier and Proper Orthogonal Decomposition spectra, the flow topology characterized by the Okubo-Weiss parameter, and the time autocorrelation function of turbulent fluctuations. Excellent agreement has also been obtained in the probability distribution function of particle displacements and in the effective turbulent diffusivity. In-depth analyses of the latent space of turbulent states, choice of hyper-parameters and alternative deep learning models for the time prediction are presented. Our results highlight the potential of AI-based surrogate models to overcome the computational challenges in turbulence simulation, which can be extended to other situations such as geophysical fluid dynamics
Leveraging Gradients for Unsupervised Accuracy Estimation under Distribution Shift
No full textInternational audienceEstimating the test performance of a model, possibly under distribution shift, without having access to the ground-truth labels is a challenging, yet very important problem for the safe deployment of machine learning algorithms in the wild. Existing works mostly rely on information from either the outputs or the extracted features of neural networks to estimate a score that correlates with the ground-truth test accuracy. In this paper, we investigate -- both empirically and theoretically -- how the information provided by the gradients can be predictive of the ground-truth test accuracy even under distribution shifts. More specifically, we use the norm of classification-layer gradients, backpropagated from the cross-entropy loss after only one gradient step over test data. Our intuition is that these gradients should be of higher magnitude when the model generalizes poorly. We provide the theoretical insights behind our approach and the key ingredients that ensure its empirical success. Extensive experiments conducted with various architectures on diverse distribution shifts demonstrate that our method significantly outperforms current state-of-the-art approaches. The code is available at https://github.com/Renchunzi-Xie/GdScor
La réception de la notion américaine d'intersectionnalité en droit français et européen
No full textInternational audienc
Réenchanter l'austérité. L'esprit ingénieur face aux réformes du service public de la météorologie
No full textInternational audienc
L'acceptabilité des pratiques de sobriété par les consommateurs : une étude exploratoire dans le domaine alimentaire
No full textInternational audienc
La réduction des déchets ménagers : une approche par la théorie des pratiques
No full textInternational audienc
Two-Phase Peridynamic Elasticity with Exponential Kernels. II: Bending, Buckling, and Vibration of Beams
No full textInternational audienc
Photo-switchable polyoxazoline additive for marine fouling release silicone coatings
No full textInternational audienceMarine fouling of shipping vessels induces ecological disasters with the transport of invasive species as well as economical costs for shipping due to frictional resistance on the surface vessels. To address this issue, fouling release coatings, without toxic biocide, gradually replaced bioactive antifouling coatings. However, the main candidates, the silicones, offer limited static fouling which can be improved by using amphiphilic additive. To render more attractive the incorporation of additive in the paint formulation, they can additionally be photosensitive allowing to repel the marine organisms under light and mitigate the fouling formation. This study presents silicone coatings using photosensitive azobenzene polyoxazoline (POx) additive able to reversibly switch from trans to cis configuration under UV/Visible (Vis) light. To avoid any additive-matrix incompatibility, that is a current issue specific to commercial hydrophilic additives, amphiphilic POx were designed. Di- and triblock POx containing hydrophobic poly(2-phenyl-2-oxazoline) blocks and hydrophilic poly(2-methyl-2-oxazoline) blocks were covalently incorporated in PDMS matrix by crosslinking via self-condensation. The impact of the macromolecular architecture of additive, UV irradiation and Vis light exposure on the surface and the mechanical properties of resulting coatings was investigated by contact angle, Atomic Force Microscopy (AFM) and nanoindentation. The reversibility of the photo-response was also evaluated. Finally, the macrofouling properties of the coated surface mediated by UV/Vis light were monitored during 120 days by real-time immersion of coatings in Atlantic Ocean and macroscopic fouling evaluation of the immersed surfaces
Black hole photon ring beyond General Relativity: an integrable parametrization
No full textInternational audienceIn recent years, the shape of the photon ring in black holes images has been argued to provide a sharp test of the Kerr hypothesis for future black hole imaging missions. In this work, we confront this proposal to beyond Kerr geometries and investigate the degeneracy in the estimations of the black hole parameters using the circlipse shape proposed by Gralla and Lupsasca. To that end, we consider a model-independent parametrization of the deviations to the Kerr black hole geometry, dubbed Kerr off shell (KOS), which preserves the fundamental symmetry structure of Kerr known as the Killing tower. Besides exhibiting a Killing tensor and thus a Carter-like constant, all the representants of this family also possess a Killing-Yano tensor and are of Petrov type D. The allowed deviations to Kerr, selected by the symmetry, are encoded in two free functions which depend respectively on the radial and polar angle coordinates. Using the symmetries, we provide an analytic study of the radial and polar motion of photon trajectories generating the critical curve, to which the subrings composing the photon ring converge. This allows us to derive a ready-to-use closed formula for the parametric critical curve in term of the free functions parametrizing the deviations to Kerr. Using this result, we confront the circlipse fitting function to four examples of Kerr-like objects and we show that it admits a high degree of degeneracy. At a given inclination, the same circlipse can fit both a Kerr black hole of a given mass and spin or a modified rotating black hole with different mass and spin parameters and a new parameter . Therefore, future tests of the Kerr hypothesis could be achieved only provided one can measure independently the mass and spin of the black hole to break this degeneracy
