Research map / Adversarial robustness and security / Certified and empirical robustness
Robust Pre-Training by Adversarial Contrastive 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 |
|---|---|---|---|
| add adversarial contrastive pre-training (ACL DS) instead of Selfie pretext pre-training | adversarial fine-tuning ResNet-18 · CIFAR-10 | RA (%): 49.83 -> 52.82 | Helped |
| add ACL (DS) pre-training to UAT++ semi-supervised training with 1% labels | UAT++ · CIFAR-10 1% labels | TA (%): 41.88 -> 75.66 | Helped |
| add ACL (DS) pre-training to UAT++ with 10% labels | UAT++ · CIFAR-10 10% labels | RA (%): 50.43 -> 51.09 | Helped |
| ACL (DS) pre-training on large model | TRADE WRN-34-10 · CIFAR-10 | RA (%): 55.46 -> 56.72 | Helped |
| add adversarial perturbations via A2A instead of standard SimCLR (ablation) | SimCLR pre-training · CIFAR-10 | TA (%): 81.57 -> 78.68 | Hurt |
| add adversarial perturbations via A2A instead of standard SimCLR (ablation) | SimCLR pre-training · CIFAR-10 | RA (%): 49.97 -> 50.12 | Helped |
| use dual stream (add standard stream to A2A) instead of A2S (ablation) | ACL · CIFAR-10 | RA (%): 52.11 -> 52.82 | Helped |
| DS over SimCLR (S2S) (ablation) | pre-training · CIFAR-10 | TA (%): 81.57 -> 82.19 | Helped |
| use dual BN (separate adversarial BN) instead of single BN (ablation) | ACL (DS) · CIFAR-10 | RA (%): 40.10 -> 52.82 | Helped |
| use dual BN instead of single BN (ablation) | ACL (A2S) · CIFAR-10 | RA (%): 35.60 -> 52.11 | Helped |
| use adversarial-branch BN rather than standard BN at fine-tuning (ablation) | ACL (DS) · CIFAR-10 | RA (%): 52.36 -> 52.82 RA; 82.46 -> 82.19 TA | Mixed |
| ACL (DS) pre-training instead of Augmix consistency regularization | supervised fine-tuning · CIFAR-10 | RA (%): 52.38 -> 52.82 | Helped |
Closest papers by meaning
- When does Contrastive Learning Preserve Adversarial Robustness from Pretraining to Finetuning? (NeurIPS 2021 · similarity 0.93)
- Adversarial Self-Supervised Contrastive Learning (NeurIPS 2020 · similarity 0.92)
- Adversarial Robustness: From Self-Supervised Pre-Training to Fine-Tuning (CVPR 2020 · similarity 0.92)
- Contrastive Self-Supervised Learning Leads to Higher Adversarial Susceptibility (AAAI 2023 · similarity 0.92)
- Weakly Supervised Contrastive Adversarial Training for Learning Robust Features from Semi-supervised Data (CVPR 2025 · similarity 0.90)
- Adversarial Training Can Provably Improve Robustness: Theoretical Analysis of Feature Learning Process Under Structured Data (ICLR 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.