ResearchTrend.AI
  • Papers
  • Communities
  • Events
  • Blog
  • Pricing
Papers
Communities
Social Events
Terms and Conditions
Pricing
Parameter LabParameter LabTwitterGitHubLinkedInBlueskyYoutube

© 2025 ResearchTrend.AI, All rights reserved.

  1. Home
  2. Papers
  3. 2312.11510
27
0

QuadAttack: A Quadratic Programming Approach to Ordered Top-K Attacks

12 December 2023
Thomas Paniagua
Ryan Grainger
Tianfu Wu
    AAML
ArXivPDFHTML
Abstract

The adversarial vulnerability of Deep Neural Networks (DNNs) has been well-known and widely concerned, often under the context of learning top-111 attacks (e.g., fooling a DNN to classify a cat image as dog). This paper shows that the concern is much more serious by learning significantly more aggressive ordered top-KKK clear-box~\footnote{ This is often referred to as white/black-box attacks in the literature. We choose to adopt neutral terminology, clear/opaque-box attacks in this paper, and omit the prefix clear-box for simplicity.} targeted attacks proposed in Adversarial Distillation. We propose a novel and rigorous quadratic programming (QP) method of learning ordered top-KKK attacks with low computing cost, dubbed as \textbf{QuadAttacKKK}. Our QuadAttacKKK directly solves the QP to satisfy the attack constraint in the feature embedding space (i.e., the input space to the final linear classifier), which thus exploits the semantics of the feature embedding space (i.e., the principle of class coherence). With the optimized feature embedding vector perturbation, it then computes the adversarial perturbation in the data space via the vanilla one-step back-propagation. In experiments, the proposed QuadAttacKKK is tested in the ImageNet-1k classification using ResNet-50, DenseNet-121, and Vision Transformers (ViT-B and DEiT-S). It successfully pushes the boundary of successful ordered top-KKK attacks from K=10K=10K=10 up to K=20K=20K=20 at a cheap budget (1×601\times 601×60) and further improves attack success rates for K=5K=5K=5 for all tested models, while retaining the performance for K=1K=1K=1.

View on arXiv
Comments on this paper