Adversarial attacks for machine learning denoisers and how to resist them.

First Authors Saiyam B. Jain
Authors Saiyam B. Jain, Shao Zongru, Sachin K. T. Veettil, Michael Hecht
Corresponding Authors Michael Hecht
Last Authors Michael Hecht
Conference Proceedings Volume Title Emerging Topics in Artificial Intelligence (ETAI) 2022
Series Title (Proceedings of SPIE ; 12204)
Conference Name SPIE Nanoscience and Engineering
Conference Location San Diego, USA
Conference Start Date 2022-08-21
Conference End Date 2022-08-26
Chapter Number 1220402
Publisher SPIE
Conference Proceedings Editors Giovanni Volpe
First Page
Last Page
Open Access false
Print Publication Date 2022-10-03
Online Publication Date 2022-10-03
Abstract Adversarial attacks rely on the instability phenomenon appearing in general for all inverse problems, e.g., image classification and reconstruction, independently of the computational scheme or method used to solve the problem. We mathematically prove and empirically show that machine learning denoisers (MLD) are not excluded. That is to prove the existence of adversarial attacks given by noise patterns making the MLD run into instability, i.e., the MLD increases the noise instead of decreasing it. We further demonstrate that adversarial retraining or classic filtering do not provide an exit strategy for this dilemma. Instead, we show that adversarial attacks can be inferred by polynomial regression. Removing the underlying inferred polynomial distribution from the total noise distribution delivers an efficient technique yielding robust MLDs that make consistent computer vision tasks such as image segmentation or classification more reliable.
Cover Image
Affiliated With Sbalzarini, Postdocs
Selected By
Acknowledged Services
Publication Status Published
Edoc Link
Sfx Link
DOI 10.1117/12.2632954
PubMed ID
WebOfScience Link
Alternative Full Text URL
Display Publisher Download Only true
Visible On MPI-CBG Website true
PDF Downloadable true
Created By sbalzari
Added Date 2022-10-12
Last Edited By thuem
Last Edited Date 2022-10-28 15:17:41.751
Library ID 8458
Document ID
Entry Complete true
eDoc Compliant true
Include in Edoc Report false
In Pure false
Ready for eDoc Export false
Author Affiliations Complete false
Project Name
Project URL
Grant ID
Funding Programme
Funding Organisation