Research map / Adversarial robustness and security / Certified and empirical robustness
MMA Training: Direct Input Space Margin Maximization through Adversarial Training
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 |
|---|---|---|---|
| Use per-example adaptive perturbation magnitudes to maximize margins rather than a fixed training magnitude | Adversarial training on CIFAR10 ℓ∞ · CIFAR10 ℓ∞ | AvgAcc: 28.28 -> 29.39 | Helped |
| Use per-example adaptive perturbation magnitudes rather than a fixed magnitude | Adversarial training on CIFAR10 ℓ∞ · CIFAR10 ℓ∞ | robust accuracy at ε=32: 0.69 -> 0.81 | Helped |
| Use per-example adaptive perturbation magnitudes rather than a fixed magnitude | Adversarial training on CIFAR10 ℓ∞ · CIFAR10 ℓ∞ | robust accuracy at ε=8: 52.92 -> 47.18 | Hurt |
| Linearly increase the fixed perturbation magnitude from 0 to 24 over 50 epochs | PGD-24 · CIFAR10 ℓ∞ | AvgAcc: 9.95 -> 26.53 | Helped |
| Linearly increase the fixed perturbation magnitude from 0 to 24 over 50 epochs | PGD-24 · CIFAR10 ℓ∞ | robust accuracy at ε=24: 9.69 -> 9.66 | Helped |
| Add a clean cross-entropy loss term to the MMA objective (ablation) | MMA training · CIFAR10 | training performance | Helped |
| Use clean-loss/MMA-loss weights (1/3, 2/3) rather than the other tested pairs (ablation) | MMA training · CIFAR10 ℓ∞ | performance | Helped |
| Train with a longer 80000-step schedule rather than 50000 steps (ablation) | WideResNet adversarial training · CIFAR10 | performance | No clear effect |
| Vary the initial learning rate among the tested values (ablation) | LeNet5 training · MNIST | performance | No clear effect |
| Vary the initial learning rate among the tested values (ablation) | WideResNet training · CIFAR10 | performance | No clear effect |
Closest papers by meaning
- Reducing Excessive Margin to Achieve a Better Accuracy vs. Robustness Trade-off (ICLR 2022 · similarity 0.90)
- Building Robust Ensembles via Margin Boosting (ICML 2022 · similarity 0.90)
- Detecting Brittle Decisions for Free: Leveraging Margin Consistency in Deep Robust Classifiers (NeurIPS 2024 · similarity 0.89)
- Implicit Bias of Adversarial Training for Deep Neural Networks (ICLR 2022 · similarity 0.88)
- Kernel Learning with Adversarial Features: Numerical Efficiency and Adaptive Regularization (NeurIPS 2025 · similarity 0.88)
- Adversarial Training Can Provably Improve Robustness: Theoretical Analysis of Feature Learning Process Under Structured Data (ICLR 2025 · 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.