Loading Now

Summary of Efficient Black-box Adversarial Attacks Via Bayesian Optimization Guided by a Function Prior, By Shuyu Cheng et al.


Efficient Black-box Adversarial Attacks via Bayesian Optimization Guided by a Function Prior

by Shuyu Cheng, Yibo Miao, Yinpeng Dong, Xiao Yang, Xiao-Shan Gao, Jun Zhu

First submitted to arxiv on: 29 May 2024

Categories

  • Main: Machine Learning (cs.LG)
  • Secondary: Artificial Intelligence (cs.AI); Cryptography and Security (cs.CR); Computer Vision and Pattern Recognition (cs.CV); Machine Learning (stat.ML)

     Abstract of paper      PDF of paper


GrooveSquid.com Paper Summaries

GrooveSquid.com’s goal is to make artificial intelligence research accessible by summarizing AI papers in simpler terms. Each summary below covers the same AI paper, written at different levels of difficulty. The medium difficulty and low difficulty versions are original summaries written by GrooveSquid.com, while the high difficulty version is the paper’s original abstract. Feel free to learn from the version that suits you best!

Summary difficulty Written by Summary
High Paper authors High Difficulty Summary
Read the original abstract here
Medium GrooveSquid.com (original content) Medium Difficulty Summary
This paper proposes Prior-guided Bayesian Optimization (P-BO), an algorithm that leverages a surrogate white-box model as a global function prior in black-box adversarial attacks. P-BO models the attack objective with a Gaussian process, initialized with the surrogate model’s loss, to generate efficient and effective adversarial examples against black-box models. Theoretical analysis reveals that the performance of P-BO can be affected by a bad prior, so an adaptive integration strategy is introduced to adjust a coefficient on the function prior to minimize regret. Experimental results demonstrate the superiority of P-BO in reducing queries and improving attack success rates compared to state-of-the-art black-box attacks.
Low GrooveSquid.com (original content) Low Difficulty Summary
P-BO is a new way to make computers do what we want, even if we don’t know how they work. It’s like having a secret map that helps us find the right path. This paper shows how P-BO can be used to trick computers into doing things they wouldn’t normally do, and it’s much better at doing this than other ways that people have tried.

Keywords

» Artificial intelligence  » Optimization