Layer-wise regularized adversarial training using layers sustainability analysis framework

doi:10.1016/j.neucom.2023.03.043

Layer-wise regularized adversarial training using layers sustainability analysis framework

2023-07-01

0

PRE

AI

M

Mohammad Mehdi Homayounpour *

M

Maryam Amirmazlaghani

DOI：10.1016/j.neucom.2023.03.043

原文链接

原文求助

摘要

En 中文

Deep neural network models are used today in various applications of artificial intelligence, the strength-ening of which, in the face of adversarial attacks is of particular importance. An appropriate solution to adversarial attacks is adversarial training, which reaches a trade-off between robustness and generaliza-tion. This paper introduces a novel framework (Layer Sustainability Analysis (LSA)) for the analysis of layer vulnerability in an arbitrary neural network in the scenario of adversarial attacks. LSA can be a help-ful toolkit to assess deep neural networks and to extend the adversarial training approaches towards improving the sustainability of model layers via layer monitoring and analysis. The LSA framework iden-tifies a list of Most Vulnerable Layers (MVL list) of the given network. The relative error, as a comparison measure, is used to evaluate representation sustainability of each layer against adversarial inputs. The proposed approach for obtaining robust neural networks to fend off adversarial attacks is based on a layer-wise regularization (LR) over LSA proposal(s) for adversarial training (AT). This means that the AT-LR procedure could be used with any benchmark adversarial attack to reduce the vulnerability of net -work layers and to improve conventional adversarial training approaches. The proposed idea performs well theoretically and experimentally for state-of-the-art multilayer perceptron and convolutional neural network architectures. Additionally, a measure named robustness and generalization score or R&G score is defined to better evaluate each adversarially trained model over a variety of significant perturbations. Compared with the AT-LR and its corresponding base adversarial training, the R&G score on Moon, MNIST, and CIFAR-10 benchmark datasets was increased by 56.52%, 75.82%, and 6.54%, respectively for more significant perturbations. The LSA framework is available and published at https://github.com/kha-looei/LSA. (c) 2023 Elsevier B.V. All rights reserved.

Keyword:

Layer Sustainability Analysis

Deep neural networks

Adversarial training

Adversarial defense

Robustness

Generalization

Layer-wise regularized adversarial training using layers sustainability analysis framework

摘要

期刊

机构

引用论文