Gradient-Informed Neural Network Statistical Robustness Estimation - Inria - Institut national de recherche en sciences et technologies du numérique
Conference Papers Year : 2023

Gradient-Informed Neural Network Statistical Robustness Estimation

Abstract

Deep neural networks are robust against random corruptions of the inputs to some extent. This global sense of safety is not sufficient in critical applications where probabilities of failure must be assessed with accuracy. Some previous works applied known statistical methods from the field of rare event analysis to classification. Yet, they use classifiers as black-box models without taking into account gradient information, readily available for deep learning models via autodifferentiation. We propose a new and highly efficient estimator of probabilities of failure dedicated to neural networks as it leverages the fast computation of gradients of the model through back-propagation.
Fichier principal
Vignette du fichier
AISTSAT2023_gradient_informed_nn_robust_estim.pdf (5.73 Mo) Télécharger le fichier
Origin Files produced by the author(s)

Dates and versions

hal-03987284 , version 1 (14-02-2023)

Identifiers

  • HAL Id : hal-03987284 , version 1

Cite

Karim Tit, Teddy Furon, Mathias Rousset. Gradient-Informed Neural Network Statistical Robustness Estimation. AISTATS 2023 - 26th International Conference on Artificial Intelligence and Statistics, Apr 2023, Valencia, Spain. ⟨hal-03987284⟩
115 View
163 Download

Share

More