Interval based verification of adversarial example free zones for neural networks – Dependency problem

Csendes, Tibor (2024) Interval based verification of adversarial example free zones for neural networks – Dependency problem. ANNALES MATHEMATICAE ET INFORMATICAE, 60. pp. 19-26. ISSN 1787-6117

Preview

Text AMI_60_from19to26.pdf - Published Version Download (498kB) | Preview

Abstract

Recent machine learning models are sensitive to adversarial input perturbation. That is, an attacker may easily mislead an otherwise wellperforming image classification system by altering some pixels. It is quite challenging to prove that a network will have correct output when changing slightly some regions of the images. This is why only a few works targeted this problem. Although there are an increasing number of studies on this field, really reliable robustness evaluation is still an open issue. We will present some theoretical results on the dependency problem of interval arithmetic what is critical in interval based verification.

Actions (login required)

Edit Item