Research map / Adversarial robustness and security / Black-box and transfer attacks
TransferBench: Benchmarking Ensemble-based Black-box Transfer Attacks
Fabio Brau, Maura Pintor, Antonio Emanuele Cinà, Raffaele Mura, Luca Scionis, Luca Oneto, Fabio Roli, Battista Biggio
Read the paperCheck your idea against 137,217 papers
AI-extracted summary. Written by AI from the paper's abstract only. Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Shows that prior evaluations of ensemble-based black-box transfer attacks are systematically biased by optimistic test scenarios, and introduces TransferBench to benchmark them under more realistic conditions.
Problem
Ensemble-based black-box transfer attacks are claimed to achieve high success rates with few target-model queries, but prior evaluations may be overly optimistic, ignoring factors such as how surrogate model choice affects transferability.
Approach
Argues that existing evaluations test ensemble transfer attacks only under overly optimistic scenarios and proposes a benchmark (TransferBench) that accounts for factors such as surrogate model choice to give a less biased assessment.
Surprising findings
- Prior evaluations of ensemble-based black-box transfer attacks are systematically biased toward overly optimistic scenarios.
Experimental setup
Benchmarks: TransferBench
Closest papers by meaning
- Rethinking Model Ensemble in Transfer-based Adversarial Attacks (ICLR 2024 · similarity 0.91)
- Enhancing Adversarial Transferability with Checkpoints of a Single Model's Training (CVPR 2025 · similarity 0.91)
- PubDef: Defending Against Transfer Attacks From Public Models (ICLR 2024 · similarity 0.90)
- Understanding Model Ensemble in Transferable Adversarial Attack (ICML 2025 · similarity 0.90)
- LRS: Enhancing Adversarial Transferability through Lipschitz Regularized Surrogate (AAAI 2024 · similarity 0.90)
- Enhancing Adversarial Transferability with Adversarial Weight Tuning (AAAI 2025 · similarity 0.89)
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.
