We gratefully acknowledge support from
the Simons Foundation and member institutions.
Full-text links:

Download:

Current browse context:

cs.LG

Change to browse by:

References & Citations

DBLP - CS Bibliography

Bookmark

(what is this?)
CiteULike logo BibSonomy logo Mendeley logo del.icio.us logo Digg logo Reddit logo ScienceWISE logo

Computer Science > Machine Learning

Title: MaskBlock: Transferable Adversarial Examples with Bayes Approach

Abstract: The transferability of adversarial examples (AEs) across diverse models is of critical importance for black-box adversarial attacks, where attackers cannot access the information about black-box models. However, crafted AEs always present poor transferability. In this paper, by regarding the transferability of AEs as generalization ability of the model, we reveal that vanilla black-box attacks craft AEs via solving a maximum likelihood estimation (MLE) problem. For MLE, the results probably are model-specific local optimum when available data is small, i.e., limiting the transferability of AEs. By contrast, we re-formulate crafting transferable AEs as the maximizing a posteriori probability estimation problem, which is an effective approach to boost the generalization of results with limited available data. Because Bayes posterior inference is commonly intractable, a simple yet effective method called MaskBlock is developed to approximately estimate. Moreover, we show that the formulated framework is a generalization version for various attack methods. Extensive experiments illustrate MaskBlock can significantly improve the transferability of crafted adversarial examples by up to about 20%.
Comments: Under Review
Subjects: Machine Learning (cs.LG); Cryptography and Security (cs.CR); Computer Vision and Pattern Recognition (cs.CV)
Cite as: arXiv:2208.06538 [cs.LG]
  (or arXiv:2208.06538v1 [cs.LG] for this version)

Submission history

From: Mingyuan Fan [view email]
[v1] Sat, 13 Aug 2022 01:20:39 GMT (3547kb,D)

Link back to: arXiv, form interface, contact.