Research map / Adversarial robustness and security / Certified and empirical robustness
Defensive Few-shot Learning
Wenbin Li, Lei Wang, Xingxing Zhang, Lei Qi, Jing Huo, Yang Gao, Jiebo Luo
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 adversarial training to DN4 (ablation) | DN4 · miniImageNet 5-way 5-shot (eps=0.01) | adversarial accuracy: 21.54 -> 38.19 | Improved |
| add episodic training to adversarially trained DN4 (ET) (ablation) | DN4+AT · miniImageNet 5-way 5-shot (eps=0.01) | adversarial accuracy: 38.19 -> 54.97 | Improved |
| add episodic training to adversarially trained DN4 (ET), clean accuracy (ablation) | DN4+AT · miniImageNet 5-way 5-shot | clean accuracy: 52.12 -> 71.54 | Improved |
| add adversarial training to episodically trained DN4 (ablation) | DN4+Ep · miniImageNet 5-way 5-shot (eps=0.01) | adversarial accuracy: 8.65 -> 54.97 | Improved |
| add episodic training to plain DN4 without AT (ablation) | DN4 · miniImageNet 5-way 5-shot (eps=0.01) | adversarial accuracy: 21.54 -> 8.65 | Hurt |
| add episodic training to plain DN4 without AT, clean accuracy (ablation) | DN4 · miniImageNet 5-way 5-shot | clean accuracy: 57.74 -> 70.39 | Improved |
| apply ET to ProtoNet+AT (ablation) | ProtoNet · miniImageNet 5-way 5-shot | clean accuracy: 54.61 -> 63.70 | Improved |
| apply ET to DeepEMD+AT (ablation) | DeepEMD · miniImageNet 5-way 5-shot | clean accuracy: 55.09 -> 60.20 | Improved |
| use DN4 classifier instead of ProtoNet within DFSL-AT | DFSL framework · miniImageNet 5-way 5-shot (eps=0.01) | adversarial accuracy: 39.05 -> 54.97 | Improved |
| add SKL prediction-wise consistency to DFSL-DN4-AT | DFSL-DN4 · miniImageNet 5-way 5-shot (eps=0.01) | adversarial accuracy: 54.97 -> 62.68 | Improved |
| add TCD feature-wise consistency to DFSL-DN4-AT | DFSL-DN4 · tieredImageNet 5-way 5-shot (eps=0.01) | adversarial accuracy: 53.73 -> 63.67 | Improved |
| use SKL instead of ALP | DFSL-DN4 · CIFAR-100 5-way 5-shot (eps=0.01) | adversarial accuracy: 40.59 -> 44.79 | Improved |
Closest papers by meaning
- Generate Universal Adversarial Perturbations for Few-Shot Learning (NeurIPS 2024 · similarity 0.87)
- Adversarially Robust Few-shot Learning via Parameter Co-distillation of Similarity and Class Concept Learners (CVPR 2024 · similarity 0.87)
- Few-Shot Adversarial Prompt Learning on Vision-Language Models (NeurIPS 2024 · similarity 0.86)
- Adversarial Robustness via Deformable Convolution with Stochasticity (ICML 2025 · similarity 0.85)
- Understanding the Robustness of Randomized Feature Defense Against Query-Based Adversarial Attacks (ICLR 2024 · similarity 0.85)
- Taxonomy Driven Fast Adversarial Training (AAAI 2024 · 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.
