Research map / Optimization / Optimizers
Min-Max Optimization without Gradients: Convergence and Applications to Black-Box Evasion and Poisoning Attacks
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 |
|---|---|---|---|
| ZO gradient estimation with q random directions in place of first-order gradients in the x-step (black-box x) | ZO-Min-Max one-sided ensemble evasion attack · Ensemble universal perturbation attack | attack loss per model-class pair: slightly worse than FO-Min-Max, but needs no model internals | Mixed |
| Increasing number of random direction vectors q from small to q>=5 (ablation) | ZO-Min-Max data poisoning attack · Poisoning attack on logistic regression | stationary gap | Helped |
| Replacing BO-based STABLEOPT with ZO-Min-Max (ZO gradient estimation and alternating GDA) | Black-box data poisoning attack · Poisoning attack on logistic regression | testing accuracy vs computation time | Helped |
| Using ZO-PGA (two-sided black-box) instead of FO-PGA for the y-step | ZO-Min-Max · Theory (convergence) | convergence error bound | Hurt |
| Using one-step PGA on w instead of exact inner maximization | ZO-Min-Max ensemble attack · Ensemble universal perturbation attack | attack loss convergence speed | Hurt |
Closest papers by meaning
- Robust and Faster Zeroth-Order Minimax Optimization: Complexity and Applications (NeurIPS 2024 · similarity 0.87)
- Efficient Black-box Adversarial Attacks via Bayesian Optimization Guided by a Function Prior (ICML 2024 · similarity 0.87)
- $\sigma$-zero: Gradient-based Optimization of $\ell_0$-norm Adversarial Examples (ICLR 2025 · similarity 0.86)
- DeepZero: Scaling Up Zeroth-Order Optimization for Deep Model Training (ICLR 2024 · similarity 0.86)
- Refining Adaptive Zeroth-Order Optimization at Ease (ICML 2025 · similarity 0.86)
- Natural Perturbations for Black-box Training of Neural Networks by Zeroth-Order Optimization (ICML 2025 · similarity 0.85)
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.