Research map / Adversarial robustness and security / Certified and empirical robustness
Self-ensemble Adversarial Training for Improved Robustness
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 |
|---|---|---|---|
| Average historical adversarial-training weights with SEAT | ResNet18 · CIFAR-10 | AA robust accuracy: 44.37±0.37 -> 51.3±0.26 | Helped |
| Average historical adversarial-training weights with SEAT | WRN-32-10 · CIFAR-10 | AA robust accuracy: 48.17±0.48 -> 55.67±0.22 | Helped |
| Average historical adversarial-training weights with SEAT | ResNet18 · CIFAR-100 | AA robust accuracy: 23.91±0.41 -> 27.87±0.24 | Helped |
| Add CutMix to SEAT (ablation) | WRN-32-10 · CIFAR-10 | AA robust accuracy: 55.67±0.22 -> 56.03±0.36 | Helped |
| Add CutMix to SEAT (ablation) | ResNet18 · CIFAR-10 | AA robust accuracy: 51.3±0.26 -> 49.1±0.44 | Hurt |
| Use a piecewise linear rather than staircase learning-rate schedule (ablation) | SEAT · CIFAR-10, ResNet18 | AA robust accuracy: 45.85 -> 51.3 | Helped |
| Use a cosine rather than staircase learning-rate schedule (ablation) | SEAT · CIFAR-10, ResNet18 | AA robust accuracy: 45.85 -> 50.48 | Helped |
| Use a cyclic rather than staircase learning-rate schedule (ablation) | SEAT · CIFAR-10, ResNet18 | AA robust accuracy: 45.85 -> 51.44 | Helped |
| Use warmup rather than staircase learning-rate schedule (ablation) | SEAT · CIFAR-10, ResNet18 | AA robust accuracy: 45.85 -> 45.82 | No clear effect |
| Replace piecewise linear learning-rate schedule with staircase schedule (ablation) | SEAT · CIFAR-10, ResNet18 | AA robust accuracy: 51.3 -> 45.85 | Hurt |
| Combine SEAT with TRADES (ablation) | SEAT · CIFAR-10, ResNet18 | AA robust accuracy: 51.3±0.26 -> 49.91±0.17 | Hurt |
| Combine SEAT with MART (ablation) | SEAT · CIFAR-10, ResNet18 | AA robust accuracy: 51.3±0.26 -> 51.7±0.35 | Helped |
Closest papers by meaning
- Ensemble Adversarial Training: Attacks and Defenses (ICLR 2018 · similarity 0.92)
- MORA: Improving Ensemble Robustness Evaluation with Model Reweighing Attack (NeurIPS 2022 · similarity 0.91)
- Improving Ensemble Robustness by Collaboratively Promoting and Demoting Adversarial Robustness (AAAI 2021 · similarity 0.89)
- Understanding and Improving Ensemble Adversarial Defense (NeurIPS 2023 · similarity 0.89)
- Rethinking Model Ensemble in Transfer-based Adversarial Attacks (ICLR 2024 · similarity 0.87)
- 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.