Research map / Adversarial robustness and security / Certified and empirical robustness
On Fast Adversarial Robustness Adaptation in Model-Agnostic Meta-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 |
|---|---|---|---|
| Apply robust regularization only during meta-update, rather than during both fine-tuning and meta-update (ablation) | Robust MAML on MiniImageNet · MiniImageNet 5-way 1-shot | RA | No clear effect |
| Use standard instead of adversarial fine-tuning at meta-testing (ablation) | RMAMLb · MiniImageNet 5-way 1-shot | RA: 23.5% vs. 22.9% | No clear effect |
| Replace RMAMLb meta-update regularization with AQ's adversarial-querying objective (ablation) | RMAMLb · MiniImageNet 5-way 1-shot | RA: about 2% | Helped |
| Replace RMAMLb with AQ (ablation) | RMAMLb · MiniImageNet 5-way 1-shot | SA: 11% | Hurt |
| Use ANIL, fine-tuning only the classification head (ablation) | RMAMLb · MiniImageNet 5-way 1-shot | RA: 22.9% -> 22.7% | Hurt |
| Use FGSM instead of multi-step attack generation (ablation) | RMAMLb · MiniImageNet 5-way 1-shot | RA: 22.9% -> 23.04% | Helped |
| Use FGSM instead of multi-step attack generation (ablation) | RMAMLb · MiniImageNet 5-way 1-shot | Time per epoch: 54min -> 44min | Helped |
| Add unlabeled data to TRADES regularization (ablation) | RMAMLb · MiniImageNet 5-way 1-shot | RA | Helped |
| Add unlabeled data to TRADES regularization (ablation) | RMAMLb · MiniImageNet 5-way 1-shot | SA | Hurt |
| Add contrastive learning using standard and adversarial views (ablation) | RMAMLb-TRADES · MiniImageNet 5-way 1-shot | RA: 25.51% -> 26.81% | Helped |
| Add contrastive learning using standard and adversarial views (ablation) | RMAMLb-TRADES · MiniImageNet 5-way 1-shot | SA: 37.1% -> 38.60% | Helped |
| Add contrastive learning using standard and adversarial views (ablation) | RMAMLb-TRADES · CIFAR-FS 1-shot 5-way | RA: 28.06% -> 29.33% | Helped |
Closest papers by meaning
- Repurposing Pretrained Models for Robust Out-of-domain Few-Shot Learning (ICLR 2021 · similarity 0.90)
- Meta-Learning with Neural Tangent Kernels (ICLR 2021 · similarity 0.89)
- Task-Robust Model-Agnostic Meta-Learning (NeurIPS 2020 · similarity 0.89)
- Sharp-MAML: Sharpness-Aware Model-Agnostic Meta Learning (ICML 2022 · similarity 0.89)
- Adversarially Robust Few-Shot Learning: A Meta-Learning Approach (NeurIPS 2020 · similarity 0.88)
- Adversarial Training Can Provably Improve Robustness: Theoretical Analysis of Feature Learning Process Under Structured Data (ICLR 2025 · 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.