Deakin University
Browse

SSIMLayer: Towards robust deep representation learning via nonlinear structural similarity

Version 2 2024-06-04, 02:22
Version 1 2020-01-21, 01:54
conference contribution
posted on 2024-06-04, 02:22 authored by A Abobakr, M Hossny, S Nahavandi
© 2019 IEEE. Adversarial examples form a major threat to incorporating machine learning (ML) models in critical applications. The existence and generalisation of these attacks have been attributed to the linear nature of ML models, deep neural network models in particular, in the high dimensional space. This paper presents a new nonlinear computational layer to the deep convolutional neural network architectures. This layer performs a set of comprehensive convolution operations that mimics the overall function of the human visual system (HVS) via focusing on learning structural information. The core of its computations is evaluating the components of the structural similarity metric (SSIM) in a setting that allows the kernels to learn to match structural information. The proposed SSIMLayer is inherently nonlinear. Experiments conducted on CIFAR-10 benchmark demonstrate that the SSIMLayer provides high learning capacity and shows more robustness against adversarial attacks.

History

Pagination

1234-1238

Location

Bari, Italy

Start date

2019-10-06

End date

2019-10-09

ISSN

1062-922X

ISBN-13

9781728145693

Language

eng

Publication classification

E1 Full written paper - refereed

Title of proceedings

SMC 2019 : Proceedings of the IEEE International Conference on Systems, Man and Cybernetics

Event

Systems, Man and Cybernetics. Conference (2019 : Bari, Italy)

Publisher

IEEE

Place of publication

Piscataway, N.J.

Usage metrics

    Research Publications

    Exports

    RefWorks
    BibTeX
    Ref. manager
    Endnote
    DataCite
    NLM
    DC