Noise Sensitivity and Stability of Deep Neural Networks for Binary Classification
Artikel i vetenskaplig tidskrift, 2023

A first step is taken towards understanding often observed non-robustness phenomena of deep neural net (DNN) classifiers. This is done from the perspective of Boolean functions by asking if certain sequences of Boolean functions represented by common DNN models are noise sensitive or noise stable, concepts defined in the Boolean function literature. Due to the natural randomness in DNN models, these concepts are extended to annealed and quenched versions. Here we sort out the relation between these definitions and investigate the properties of two standard DNN architectures, the fully connected and convolutional models, when initiated with Gaussian weights.

Deep neural networks

Feed forward neural networks

Boolean functions

Noise sensitivity

Noise stability

Författare

Johan Jonasson

Chalmers, Matematiska vetenskaper, Analys och sannolikhetsteori

Jeffrey Steif

Chalmers, Matematiska vetenskaper, Analys och sannolikhetsteori

Olof Zetterqvist

Chalmers, Matematiska vetenskaper, Tillämpad matematik och statistik

Stochastic Processes and their Applications

0304-4149 (ISSN)

Vol. 165 130-167

Interagerande partikelsystem, cellulära automater, kvasilokalitet och färgrepresentationer.

Vetenskapsrådet (VR) (2020-03763), 2021-01-01 -- 2024-12-31.

Ämneskategorier

Annan matematik

DOI

10.1016/j.spa.2023.08.003

Mer information

Senast uppdaterat

2024-04-23