Research map / Adversarial robustness and security / Certified and empirical robustness
Adversarial Distributional Training for Robust Deep Learning
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 |
|---|---|---|---|
| Train on worst-case adversarial distributions instead of single attack examples | ADT_EXP vs AT_PGD · CIFAR-10, white-box attacks | overall robust accuracy: 48.26% -> 50.56% | Helped |
| Train on worst-case adversarial distributions instead of single attack examples | ADT_EXP vs AT_PGD · CIFAR-100, white-box attacks | overall robust accuracy: 24.49% -> 27.13% | Helped |
| Train on worst-case adversarial distributions instead of single attack examples | ADT_EXP vs AT_PGD · SVHN, white-box attacks | overall robust accuracy: 73.38% -> 75.55% | Helped |
| Use amortized explicit distributions rather than per-example explicit distributions (ablation) | ADT_EXP-AM vs ADT_EXP · CIFAR-10, white-box attacks | overall robust accuracy: 50.56% -> 50.04% | Hurt |
| Use amortized implicit distributions rather than per-example explicit distributions (ablation) | ADT_IMP-AM vs ADT_EXP · CIFAR-10, white-box attacks | overall robust accuracy: 50.56% -> 49.81% | Hurt |
| Use amortized explicit distributions rather than per-example explicit distributions (ablation) | ADT_EXP-AM vs ADT_EXP · CIFAR-100, white-box attacks | overall robust accuracy: 27.13% -> 26.87% | Hurt |
| Use amortized implicit distributions rather than per-example explicit distributions (ablation) | ADT_IMP-AM vs ADT_EXP · CIFAR-100, white-box attacks | overall robust accuracy: 27.13% -> 26.80% | Hurt |
| Use amortized explicit distributions rather than per-example explicit distributions (ablation) | ADT_EXP-AM vs ADT_EXP · SVHN, white-box attacks | overall robust accuracy: 75.55% -> 75.00% | Hurt |
| Use amortized implicit distributions rather than per-example explicit distributions (ablation) | ADT_IMP-AM vs ADT_EXP · SVHN, white-box attacks | overall robust accuracy: 75.55% -> 74.13% | Hurt |
| Integrate ADT with TRADES loss | ADT_EXP vs TRADES · CIFAR-10, beta 1.0 | overall robust accuracy: 47.75% -> 49.05% | Helped |
| Integrate ADT with TRADES loss | ADT_EXP-AM vs TRADES · CIFAR-10, beta 1.0 | overall robust accuracy: 47.75% -> 50.78% | Helped |
| Integrate ADT with TRADES loss | ADT_EXP vs TRADES · CIFAR-10, beta 6.0 | overall robust accuracy: 52.64% -> 54.21% | Helped |
Closest papers by meaning
- A Unified Wasserstein Distributional Robustness Framework for Adversarial Training (ICLR 2022 · similarity 0.92)
- Adversarial Robustness without Adversarial Training: A Teacher-Guided Curriculum Learning Approach (NeurIPS 2021 · similarity 0.91)
- Adversarial Training Can Provably Improve Robustness: Theoretical Analysis of Feature Learning Process Under Structured Data (ICLR 2025 · similarity 0.90)
- Revisiting Adversarial Training Under Long-Tailed Distributions (CVPR 2024 · similarity 0.89)
- TAET: Two-Stage Adversarial Equalization Training on Long-Tailed Distributions (CVPR 2025 · similarity 0.89)
- Enhancing the Adversarial Robustness via Manifold Projection (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.