Research map / Adversarial robustness and security / Black-box and transfer attacks
On the Design of Black-Box Adversarial Examples by Leveraging Gradient-Free Optimization and Operator Splitting Method
Pu Zhao, Sijia Liu, Pin-Yu Chen, Nghia Hoang, Kaidi Xu, Bhavya Kailkhura, Xue Lin
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| use ZO-ADMM with linearized single-step gradient estimation instead of ZOO | black-box l2 attack · MNIST | query count on initial success: 12,161 -> 493.6 | Improved |
| use ZO-ADMM instead of ZOO | black-box l2 attack · CIFAR-10 | query count on initial success: 9982 -> 421 | Improved |
| use BO-ADMM instead of RGE-based ZO-ADMM | score-based ADMM attack · MNIST | query count / l2 distortion: 493.6 -> 52.1 queries; l2 1.975 -> 7.7 | Mixed |
| use BO-ADMM instead of RGE-based ZO-ADMM | score-based ADMM attack · CIFAR-10 | attack success rate: 98.7 -> 84.1 | Hurt |
| initialize ZO-ADMM with BO (BO-ZO-ADMM) | ZO-ADMM · MNIST | queries to same l2 distortion: 493 -> 380 | Improved |
| initialize ZO-ADMM with BO (BO-ZO-ADMM) | ZO-ADMM · CIFAR-10 | queries to same l2 distortion: 421 -> 320 | Improved |
| use decision-based ZO-ADMM instead of boundary attack | hard-label attack · MNIST | query count: 25,328 -> 7,603 | Improved |
| use decision-based ZO-ADMM instead of boundary attack | hard-label attack · CIFAR-10 | query count: 16,720 -> 6,213 | Improved |
| use ZO-ADMM instead of Bandits_TD | score-based attack · ImageNet untargeted | query count on initial success: 1259 -> 891 | Improved |
| use ZO-ADMM instead of AutoZoom | black-box attack · ImageNet | number of queries: 43547 -> 16390 | Improved |
| use ZO-ADMM instead of query-limited attack | score-based attack · ImageNet targeted | query count on initial success: 34128 -> 16058 | Improved |
| switch distortion term from l2 to l0 in ZO-ADMM | ZO-ADMM · MNIST | l0 distortion: 483 -> 18.5 | Improved |
Closest papers by meaning
- Towards Query-Efficient Black-Box Adversary with Zeroth-Order Natural Gradient Descent (AAAI 2020 · similarity 0.90)
- How to Robustify Black-Box ML Models? A Zeroth-Order Optimization Perspective (ICLR 2022 · similarity 0.90)
- Prior Convictions: Black-box Adversarial Attacks with Bandits and Priors (ICLR 2019 · similarity 0.89)
- Efficient Black-box Adversarial Attacks via Bayesian Optimization Guided by a Function Prior (ICML 2024 · similarity 0.89)
- Understanding the Robustness of Randomized Feature Defense Against Query-Based Adversarial Attacks (ICLR 2024 · similarity 0.89)
- BRUSLEATTACK: A QUERY-EFFICIENT SCORE- BASED BLACK-BOX SPARSE ADVERSARIAL ATTACK (ICLR 2024 · similarity 0.87)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
