Research map / Adversarial robustness and security / Certified and empirical robustness
Efficient Robust Training via Backward Smoothing
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 backward smoothing instead of single-step random initialization | fast robust training · CIFAR-10 | robust accuracy (%): 46.30 -> 52.50 | Helped |
| use backward smoothing instead of Fast TRADES | fast robust training · CIFAR-10 | robust accuracy (%): 46.98 -> 52.50 | Helped |
| use backward smoothing instead of TRADES | robust training · CIFAR-10 | robust accuracy (%): 52.74 -> 52.50 | No clear effect |
| use backward smoothing instead of TRADES | robust training · CIFAR-100 | robust accuracy (%): 29.41 -> 30.50 | Helped |
| use backward smoothing instead of TRADES | robust training · Tiny ImageNet | robust accuracy (%): 21.04 -> 22.32 | Helped |
| add the backward-smoothing initialization | Fast AT on CIFAR-10 · CIFAR-10 | training time (m): 82 -> 164 | Hurt |
| add the backward-smoothing initialization | Fast AT on CIFAR-10 · CIFAR-10 | robust accuracy (%): 46.30 -> 52.50 | Helped |
| remove the gradient-based attack and keep only backward-smoothing initialization (ablation) | Backward Smoothing · CIFAR-10 | robust accuracy (%): 52.50 -> 39.26 | Hurt |
| remove the gradient-based attack and keep only backward-smoothing initialization (ablation) | Backward Smoothing · CIFAR-100 | robust accuracy (%): 30.50 -> 18.51 | Hurt |
| use 2 random points instead of 1 (ablation) | Backward Smoothing · CIFAR-10 | robust accuracy (%): 52.50 -> 52.67 | No clear effect |
| increase random points from 1 to 2 (ablation) | Backward Smoothing · CIFAR-10 | training time (m): 164 -> 204 | Hurt |
| set gamma to 0.1 instead of 1.0 (ablation) | Backward Smoothing · CIFAR-10 | robust accuracy (%): 52.50 -> 52.13 | Hurt |
Closest papers by meaning
- Treatment of Statistical Estimation Problems in Randomized Smoothing for Adversarial Robustness (NeurIPS 2024 · similarity 0.90)
- Provably Cost-Sensitive Adversarial Defense via Randomized Smoothing (ICML 2025 · similarity 0.89)
- Adversarial Training Can Provably Improve Robustness: Theoretical Analysis of Feature Learning Process Under Structured Data (ICLR 2025 · similarity 0.89)
- Stability and Generalization of Adversarial Training for Shallow Neural Networks with Smooth Activation (NeurIPS 2024 · similarity 0.89)
- Adversarial Initialization with Universal Adversarial Perturbation: A New Approach to Fast Adversarial Training (AAAI 2024 · similarity 0.89)
- Kernel Learning with Adversarial Features: Numerical Efficiency and Adaptive Regularization (NeurIPS 2025 · similarity 0.88)
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.