Research map / Label-efficient and robust learning / Zero-/few-shot and meta-learning
Learning to Learn from APIs: Black-Box Data-Free Meta-Learning
Zixuan Hu, Li Shen, Zhenyi Wang, Baoyuan Wu, Chun Yuan, Dacheng Tao
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 BiDf-MKD bi-level meta knowledge distillation on top of task memory replay (ablation) | vanilla meta-learning from synthetic data · API-SS CIFAR-FS 5-way | accuracy (1-shot): +1.96 | Improved |
| add boundary query set recovery (ablation) | BiDf-MKD · API-SS CIFAR-FS 5-way | accuracy (1-shot): +0.89 | Improved |
| add boundary query set recovery (ablation) | BiDf-MKD · API-SS CIFAR-FS 5-way | accuracy (5-shot): +0.79 | Improved |
| use zero-order gradient estimator instead of first-order gradients for task recovery (ablation) | BiDf-MKD black-box setting · API-SS CIFAR-FS 5-way | accuracy (5-shot): 51.16 vs 47.58 | Hurt |
| increase number of APIs from 1 to 100 (ablation) | BiDf-MKD · API-SS CIFAR-FS 5-way | accuracy (1-shot): 30.76 to 35.48 | Improved |
| increase number of zero-order queries per API (q) from 10 to 200 (ablation) | BiDf-MKD · API-SS CIFAR-FS 5-way | accuracy (1-shot): 31.39 at q=10 to 36.32 at q=200 | Improved |
| vary lambda_Q between 0.1 and 10 in boundary query recovery (ablation) | BiDf-MKD · API-SS CIFAR-FS 5-way | accuracy (1-shot): 35.06 to 35.32 | No clear effect |
| proposed BiDf-MKD with task memory replay vs. baselines on API-SS | meta-initialization from black-box APIs · API-SS CIFAR-FS 5-way | accuracy (1-shot): 35.48 vs best baseline 24.24 | Improved |
Closest papers by meaning
- Task-Distributionally Robust Data-Free Meta-Learning (TPAMI 2025 · similarity 0.87)
- Memory-Reduced Meta-Learning with Guaranteed Convergence (AAAI 2025 · similarity 0.85)
- Task Groupings Regularization: Data-Free Meta-Learning with Heterogeneous Pre-trained Models (ICML 2024 · similarity 0.85)
- Any-Way Meta Learning (AAAI 2024 · similarity 0.85)
- Learning to Learn with Contrastive Meta-Objective (NeurIPS 2025 · similarity 0.85)
- Provable Meta-Learning with Low-Rank Adaptations (NeurIPS 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.
