Výsledky vyhledávání - "GARÈS, VALÉRIE"

Report

Analysis and Comparison of Two-Level KFAC Methods for Training Deep Neural Networks

Autor: Koroko, Abdoulaye, Anciaux-Sedrakian, Ani, Gharbia, Ibtihel Ben, Garès, Valérie, Haddou, Mounir, Tran, Quang Huy

As a second-order method, the Natural Gradient Descent (NGD) has the ability to accelerate training of neural networks. However, due to the prohibitive computational and memory costs of computing and inverting the Fisher Information Matrix (FIM), eff

Externí odkaz: http://arxiv.org/abs/2303.18083

Zobrazit plný text záznamu

Report

S-estimation in Linear Models with Structured Covariance Matrices

Autor: Lopuhaä, Hendrik Paul, Gares, Valerie, Ruiz-Gazen, Anne

We provide a unified approach to S-estimation in balanced linear models with structured covariance matrices. Of main interest are S-estimators for linear mixed effects models, but our approach also includes S-estimators in several other standard mult

Externí odkaz: http://arxiv.org/abs/2208.01939

Zobrazit plný text záznamu

Report

Efficient Approximations of the Fisher Matrix in Neural Networks using Kronecker Product Singular Value Decomposition

Autor: Koroko, Abdoulaye, Anciaux-Sedrakian, Ani, Gharbia, Ibtihel Ben, Garès, Valérie, Haddou, Mounir, Tran, Quang Huy

Several studies have shown the ability of natural gradient descent to minimize the objective function more efficiently than ordinary gradient descent based methods. However, the bottleneck of this approach for training deep neural networks lies in th

Externí odkaz: http://arxiv.org/abs/2201.10285

Zobrazit plný text záznamu

Report

Closed-form variance estimators for weighted and stratified dose-response function estimators using generalized propensity score

Autor: Garès, Valérie, Chauvet, Guillaume, Hajage, David

Propensity score methods are widely used in observational studies for evaluating marginal treatment effects. The generalized propensity score (GPS) is an extension of the propensity score framework, historically developed in the case of binary exposu

Externí odkaz: http://arxiv.org/abs/2007.02552

Zobrazit plný text záznamu

Akademický článek

Efficient approximations of the fisher matrix in neural networks using kronecker product singular value decomposition

Autor: Koroko Abdoulaye, Anciaux-Sedrakian Ani, Gharbia Ibtihel Ben, Garès Valérie, Haddou Mounir, Tran Quang Huy

Publikováno v: ESAIM: Proceedings and Surveys, Vol 73, Pp 218-237 (2023)

We design four novel approximations of the Fisher Information Matrix (FIM) that plays a central role in natural gradient descent methods for neural networks. The newly proposed approximations are aimed at improving Martens and Grosse’s Kronecker-fa

Externí odkaz: https://doaj.org/article/8a4fc72eae9e421f84c52e8da9df936d