Research map / Label-efficient and robust learning / Zero-/few-shot and meta-learning
Meta-Learning With Task-Adaptive Loss Function for Few-Shot Learning
Sungyong Baik, Janghoon Choi, Heewon Kim, Dohee Cho, Jaesik Min, Kyoung Mu Lee
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 cross entropy with a meta-learned loss in the inner loop (ablation) | MAML 4-CONV · miniImageNet 5-way 5-shot ablation | accuracy (%): 64.99 -> 67.42 | Improved |
| add cross entropy alongside the learned loss (ablation) | MAML with learned loss · miniImageNet 5-way 5-shot ablation | accuracy (%): 67.42 -> 67.86 | No clear effect |
| make the learned loss task-adaptive via affine transformation (ablation) | MAML with learned loss · miniImageNet 5-way 5-shot ablation | accuracy (%): 67.42 -> 68.56 | Improved |
| add semi-supervised inner loop using unlabeled query examples (ablation) | MAML with fixed learned loss · miniImageNet 5-way 5-shot ablation | accuracy (%): 67.42 -> 68.24 | Improved |
| add both task-adaptivity and semi-supervised inner loop (full MeTAL) (ablation) | MAML with fixed learned loss · miniImageNet 5-way 5-shot ablation | accuracy (%): 67.42 -> 70.52 | Improved |
| add base learner weights to the task state (ablation) | MeTAL with CE-only task state · miniImageNet 5-way 5-shot ablation | accuracy (%): 67.86 -> 68.66 | Improved |
| add base learner predictions to the task state (ablation) | MeTAL with CE-only task state · miniImageNet 5-way 5-shot ablation | accuracy (%): 67.86 -> 67.94 | No clear effect |
| add both weights and predictions to the task state (ablation) | MeTAL with CE-only task state · miniImageNet 5-way 5-shot ablation | accuracy (%): 67.86 -> 70.52 | Improved |
| apply MeTAL to MAML (4-CONV) | MAML · miniImageNet 5-way 5-shot | accuracy (%): 64.99 -> 70.52 | Improved |
| apply MeTAL to MAML++ (4-CONV) | MAML++ · miniImageNet 5-way 5-shot | accuracy (%): 68.32 -> 72.89 | Improved |
| apply MeTAL to ALFA + MAML (4-CONV) | ALFA + MAML · miniImageNet 5-way 5-shot | accuracy (%): 69.12 -> 74.10 | Improved |
| apply MeTAL to MAML (ResNet12) | MAML · miniImageNet 5-way 1-shot | accuracy (%): 58.60 -> 59.64 | Improved |
Closest papers by meaning
- Learning to Learn Task-Adaptive Hyperparameters for Few-Shot Learning (TPAMI 2023 · similarity 0.92)
- Learning to Forget for Meta-Learning via Task-and-Layer-Wise Attenuation (TPAMI 2021 · similarity 0.92)
- Task Agnostic Meta-Learning for Few-Shot Learning (CVPR 2019 · similarity 0.91)
- A Nested Bi-level Optimization Framework for Robust Few Shot Learning (AAAI 2022 · similarity 0.91)
- Probabilistic Model-Agnostic Meta-Learning (NeurIPS 2018 · similarity 0.90)
- BOIL: Towards Representation Change for Few-shot Learning (ICLR 2021 · similarity 0.90)
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.
