20
IRUS Total
Downloads

Universal adversarial robustness of texture and shape-biased models

File Description SizeFormat 
1911.10364v4.pdfAccepted version1.82 MBAdobe PDFView/Open
Title: Universal adversarial robustness of texture and shape-biased models
Authors: Co, KT
Muñoz-González, L
Kanthan, L
Glocker, B
Lupu, EC
Item Type: Conference Paper
Abstract: Increasing shape-bias in deep neural networks has been shown to improve robustness to common corruptions and noise. In this paper we analyze the adversarial robustness of texture and shape-biased models to Universal Adversarial Perturbations (UAPs). We use UAPs to evaluate the robustness of DNN models with varying degrees of shape-based training. We find that shape-biased models do not markedly improve adversarial robustness, and we show that ensembles of texture and shape-biased models can improve universal adversarial robustness while maintaining strong performance.
Issue Date: 23-Aug-2021
Date of Acceptance: 20-May-2021
URI: http://hdl.handle.net/10044/1/91993
DOI: 10.1109/ICIP42928.2021.9506325
Journal / Book Title: 2021 IEEE International Conference on Image Processing (ICIP)
Copyright Statement: © 2021 IEEE. Personal use of this material is permitted. Permission from IEEE must be obtained for all other uses, in any current or future media, including reprinting/republishing this material for advertising or promotional purposes, creating new collective works, for resale or redistribution to servers or lists, or reuse of any copyrighted component of this work in other works.
Conference Name: IEEE International Conference on Image Processing (ICIP)
Keywords: cs.CV
cs.CV
cs.CV
cs.CV
Notes: Code available at: https://github.com/kenny-co/sgd-uap-torch
Publication Status: Published
Start Date: 2021-09-19
Finish Date: 2021-09-22
Conference Place: Anchorage, AK, USA
Appears in Collections:Computing