Research map / Adversarial robustness and security / Certified and empirical robustness
Enhancing Adversarial Contrastive Learning via Adversarial Invariant Regularization
Xilie Xu, Jingfeng Zhang, Feng Liu, Masashi Sugiyama, Mohan Kankanhalli
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 AIR regularization to ACL | ACL · CIFAR-10 SLF | standard accuracy SA (%): 78.27 -> 80.03 | Improved |
| add AIR regularization to DynACL | DynACL · CIFAR-10 ALF | standard accuracy SA (%): 72.90 -> 77.42 | Improved |
| add AIR regularization to DynACL | DynACL · CIFAR-100 ALF | standard accuracy SA (%): 43.58 -> 45.63 | Improved |
| add AIR regularization to DynACL | DynACL · STL-10 SLF | robust accuracy AA (%): 46.49 -> 47.66 | Improved |
| add AIR to ACL, cross-task transfer | ACL · CIFAR-10 -> CIFAR-100 ALF | standard accuracy SA (%): 28.58 -> 36.10 | Improved |
| add AIR to ACL, cross-task transfer | ACL · CIFAR-10 -> CIFAR-100 ALF | robust accuracy AA (%): 11.09 -> 13.30 | Improved |
| add AIR to ACL, common corruptions | ACL · CIFAR-10-C SLF CS-1 | accuracy under common corruption (%): 76.57 -> 78.55 | Improved |
| add AIR to ACL, cross-task corruptions | ACL · CIFAR-10 -> CIFAR-100-C SLF CS-1 | accuracy under common corruption (%): 31.39 -> 37.82 | Improved |
| add AIR to DynACL after LP-AFF post-processing | DynACL++ · CIFAR-10 SLF | standard accuracy SA (%): 79.81 -> 81.80 | Improved |
| add AIR to DynACL on high-resolution Imagenette | DynACL · Imagenette SLF ResNet-50 | standard accuracy SA (%): 80.74 -> 81.66 | Improved |
| add adversarial AIR term only (lambda1=0, lambda2=0.5) (ablation) | ACL-AIR · CIFAR-10 SLF | robust accuracy AA (%): 37.39 -> 38.70 | Improved |
| add SIR term only (lambda1=0.5, lambda2=0) (ablation) | ACL-AIR · CIFAR-10 SLF | robust accuracy AA (%): 37.39 -> 37.51 | No clear effect |
Closest papers by meaning
- Robust Pre-Training by Adversarial Contrastive Learning (NeurIPS 2020 · similarity 0.90)
- Efficient Adversarial Contrastive Learning via Robustness-Aware Coreset Selection (NeurIPS 2023 · similarity 0.89)
- When does Contrastive Learning Preserve Adversarial Robustness from Pretraining to Finetuning? (NeurIPS 2021 · similarity 0.88)
- Decoupled Adversarial Contrastive Learning for Self-Supervised Adversarial Robustness (ECCV 2022 · similarity 0.87)
- Generalization Bounds for Adversarial Contrastive Learning (JMLR 2023 · similarity 0.87)
- Rethinking Invariance Regularization in Adversarial Training to Improve Robustness-Accuracy Trade-off (ICLR 2025 · similarity 0.85)
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.
