Research map / Adversarial robustness and security / Certified and empirical robustness
Mitigating Adversarial Vulnerability through Causal Parameter Estimation by Adversarial Double Machine Learning
Byung-Kwan Lee, Junho Kim, Yong Man Ro
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (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 |
|---|---|---|---|
| add ADML fine-tuning to adversarial training defenses | AT/TRADES/MART/AWP · CIFAR-10 AutoAttack (VGG-16) | robust accuracy: 46.3 -> 54.5 | Improved |
| add ADML to AWP | AWP · CIFAR-100 AutoAttack (WRN-70-10) | robust accuracy: 30.0 -> 37.6 | Improved |
| add ADML to Transformers | AT on ViT-S/16 · Tiny-ImageNet AutoAttack | robust accuracy: 31.8 -> 42.6 | Improved |
| add ADML to defenses on CIFAR-100/Tiny-ImageNet | AT on VGG-16 · CIFAR-100 clean accuracy | clean accuracy: 53.9 -> 52.2 | Hurt |
| add sample-splitting plus cross-fitting (ablation) | ADML · CIFAR-10 AutoAttack (VGG-16) | robust accuracy: 49.4 -> 54.4 | Improved |
| use only worst examples as treatments instead of worst and non-worst (ablation) | ADML · CIFAR-10 AutoAttack (VGG-16) | robust accuracy: 48.8 -> 54.4 | Improved |
| use non-worst examples only instead of worst only (ablation) | ADML · CIFAR-10 AutoAttack (VGG-16) | robust accuracy: 54.4 -> 45.8 | Hurt |
| add sample-splitting plus cross-fitting (ablation) | ADML · Tiny-ImageNet AutoAttack (VGG-16) | robust accuracy: 26.4 -> 32.2 | Improved |
| add ADML to TRADES with synthetic DDPM/EDM data | PORT/Gowal/HAT/SCORE/Wang · CIFAR-10 AutoAttack | robust accuracy: 60.6 -> 66.4 | Improved |
| add ADML to reduce class-wise vulnerability | AT · bottom-k classes CIFAR-10 | relative causal parameter (%): rho_Avg 67.33 (ResNet-18) | Improved |
Closest papers by meaning
- Adversarial Robustness Through the Lens of Causality (ICLR 2022 · similarity 0.91)
- Demystifying Causal Features on Adversarial Examples and Causal Inoculation for Robust Network by Adversarial Instrumental Variable Regression (CVPR 2023 · similarity 0.90)
- On the Vulnerability of Adversarially Trained Models Against Two-faced Attacks (ICLR 2024 · similarity 0.89)
- Adversarial Training Can Provably Improve Robustness: Theoretical Analysis of Feature Learning Process Under Structured Data (ICLR 2025 · similarity 0.88)
- Where and How to Attack? A Causality-Inspired Recipe for Generating Counterfactual Adversarial Examples (AAAI 2024 · similarity 0.87)
- Layer-Aware Analysis of Catastrophic Overfitting: Revealing the Pseudo-Robust Shortcut Dependency (ICML 2024 · 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.
