Research map / Label-efficient and robust learning / Zero-/few-shot and meta-learning
Weight-Space Learning for Certifiable Few-shot Transfer Learning
Fady Rezk, Royson Lee, Henry Gouk, Timothy Hospedales, Minyoung Kim
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 |
|---|---|---|---|
| replace continuous fine-tuning with diffusion-sampled finite adapter set (STEEL) | SGD / MeZO · LaMP-2 | non-vacuous episodes: 0.00% -> 65.12% | Improved |
| use STEEL instead of LoRA-Hub | LoRA-Hub · LaMP-5 | non-vacuous episodes: 63.84 -> 99.16 | Improved |
| use STEEL instead of LoRA-Hub | LoRA-Hub · LaMP-3 | non-vacuous episodes: 10.88% -> 55.80% | Improved |
| use STEEL instead of SGD | SGD · LaMP-2 | accuracy: 63.25% -> 63.74% | No clear effect |
| use STEEL instead of SGD | SGD · Flowers (16-shot 5-way) | average accuracy: 95.90% -> 84.90% | Hurt |
| use STEEL instead of MetaPB | MetaPB · FGVC-Aircraft (16-shot) | non-vacuous ratio: 85.00% -> 97.50% | Improved |
| use STEEL instead of MetaPB | MetaPB · Flowers (16-shot) | non-vacuous ratio: 65.00% -> 100.00% | Improved |
| use diffusion sampler instead of raw model zoo (ablation) | STEEL · vision few-shot | accuracy | Improved |
| increase shots to 128 on iNaturalist birds | STEEL · iNaturalist birds | certified risk gap: 6% above empirical test error at 64-128 shots | Improved |
| increase shot count from 1 to 4+ | STEEL · vision few-shot | median bound | Improved |
Closest papers by meaning
- PEFTDiff: Diffusion-Guided Transferability Estimation for Parameter-Efficient Fine-Tuning (ICCV 2025 · similarity 0.85)
- Adapted Deep Embeddings: A Synthesis of Methods for k-Shot Inductive Transfer Learning (NeurIPS 2018 · similarity 0.85)
- Few-Shot Parameter-Efficient Fine-Tuning is Better and Cheaper than In-Context Learning (NeurIPS 2022 · similarity 0.85)
- Improving Task-Specific Generalization in Few-Shot Learning via Adaptive Vicinal Risk Minimization (NeurIPS 2022 · similarity 0.85)
- Strong Baselines for Parameter-Efficient Few-Shot Fine-Tuning (AAAI 2024 · similarity 0.84)
- PACE: Marrying generalization in PArameter-efficient fine-tuning with Consistency rEgularization (NeurIPS 2024 · similarity 0.83)
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.
