Research map / Label-efficient and robust learning / Zero-/few-shot and meta-learning
Adversarial Task Up-sampling for Meta-learning
Yichen Wu, Long-Kai Huang, Ying Wei
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 loss to task up-sampling (ATU vs TU) (ablation) | MAML · Sinusoidal regression 10-shot | MSE: 0.84 -> 0.70 | Improved |
| Add adversarial loss to task up-sampling (ATU vs TU) (ablation) | MAML · Sinusoidal regression 30-shot | MSE: 0.47 -> 0.42 | Improved |
| Add adversarial loss to task up-sampling (ATU vs TU) (ablation) | MetaSGD · Sinusoidal regression 10-shot | MSE: 0.54 -> 0.49 | Improved |
| Add adversarial loss to task up-sampling (ATU vs TU) (ablation) | MAML · ISIC 5-shot | accuracy: 73.97% -> 74.50% | Improved |
| Add adversarial loss to task up-sampling (ATU vs TU) (ablation) | MAML · miniImagenet-S 1-shot | accuracy: 42.16% -> 42.60% | Improved |
| Add adversarial loss in cross-domain regression, MetaSGD phase shift (ablation) | MetaSGD · Cross-domain sinusoidal phase [-pi,0] | MSE: 2.04 ± 0.46 -> 2.53 ± 0.62 | Hurt |
| Use ATU task augmentation instead of MLTI | MAML · ISIC 5-shot | accuracy: 70.69% -> 74.50% | Improved |
| Use ATU task augmentation instead of MLTI | MAML · miniImagenet-S 1-shot | accuracy: 41.58% -> 42.60% | Improved |
| Add ATU to MAML versus plain MAML | MAML · miniImagenet-S 1-shot | accuracy: 38.27% -> 42.60% | Improved |
| Add ATU to MetaSGD instead of MLTI | MetaSGD · miniImagenet-S 1-shot | accuracy: 39.58% -> 40.52% | Improved |
| Add ATU to ANIL instead of MLTI | ANIL · miniImagenet-S 1-shot | accuracy: 39.15% -> 39.27% | Improved |
| Add ATU to MAML in cross-domain adaptation mini-S to Derm-S | MAML · mini-S to Derm-S 1-shot | accuracy: 34.46% -> 36.86% | Improved |
Closest papers by meaning
- Adaptive Task Sampling for Meta-Learning (ECCV 2020 · similarity 0.88)
- Towards Task Sampler Learning for Meta-Learning (IJCV 2024 · similarity 0.87)
- Meta-Learning without Memorization (ICLR 2020 · similarity 0.86)
- Model-based Adversarial Meta-Reinforcement Learning (NeurIPS 2020 · similarity 0.86)
- Set-based Meta-Interpolation for Few-Task Meta-Learning (NeurIPS 2022 · similarity 0.86)
- A Lazy Approach to Long-Horizon Gradient-Based Meta-Learning (ICCV 2021 · similarity 0.86)
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.
