Research map / Adversarial robustness and security / Certified and empirical robustness
MALT Powers Up Adversarial Attacks
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's full text. Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Proposes MALT, an adversarial targeting method that selects target classes using local almost-linearity assumptions, and reports it beats AutoAttack on CIFAR-100 and ImageNet.
Problem
Adversarial attacks on multi-class classifiers pick target classes naively from classifier confidence levels, which may be suboptimal.
Approach
Target selection based on local almost-linearity of the classifier at mesoscopic scale (MALT), replacing confidence-based target choice in attacks.
Key results
- MALT reportedly outperforms the state-of-the-art AutoAttack on CIFAR-100 and ImageNet (the abstract gives no specific numbers).
Limitations
- The provided text is the ICML formatting template rather than the paper body, so method details and quantitative results are unavailable.
- Evaluation details beyond the abstract are not verifiable from the given text.
Experimental setup
Datasets: CIFAR-100 ImageNet
Benchmarks: CIFAR-100 ImageNet
Metrics: attack success rate
Baselines: AutoAttack
Builds on: AutoAttack
Closest papers by meaning
- Fast and Reliable Evaluation of Adversarial Robustness with Minimum-Margin Attack (ICML 2022 · similarity 0.87)
- $\sigma$-zero: Gradient-based Optimization of $\ell_0$-norm Adversarial Examples (ICLR 2025 · similarity 0.85)
- AttackBench: Evaluating Gradient-based Attacks for Adversarial Examples (AAAI 2025 · similarity 0.85)
- SuperDeepFool: a new fast and accurate minimal adversarial attack (NeurIPS 2024 · similarity 0.84)
- Towards Million-Scale Adversarial Robustness Evaluation With Stronger Individual Attacks (CVPR 2025 · similarity 0.84)
- Efficient Robustness Evaluation via Constraint Relaxation (AAAI 2025 · similarity 0.83)
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.