Research map / Label-efficient and robust learning / Zero-/few-shot and meta-learning
Contextualizing Meta-Learning via Learning to Decompose
Han-Jia Ye, Da-Wei Zhou, Lanqing Hong, Zhenguo Li, Xiu-Shen Wei, De‐Chuan Zhan
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 LeadNet latent-attribute decomposition instead of ProtoNet-style single embedding | ProtoNet · Colored-Kuzushiji DrawClass std=0.001 1-shot 5-way | accuracy: 77.13 -> 84.43 | Improved |
| add LeadNet instead of MetaOptNet on cross-domain MiniImageNet to CUB | MetaOptNet · MiniImageNet to CUB cross-domain 5-shot 5-way | accuracy: 64.98 -> 67.89 | Improved |
| add LeadNet instead of MetaOptNet on cross-domain 1-shot | MetaOptNet · MiniImageNet to CUB cross-domain 1-shot 5-way | accuracy: 44.79 -> 44.82 | No clear effect |
| add latent attribute decomposition versus plain supervised 16-way classifier on digit attribute (ablation) | SL · Colored MNIST digit attribute | accuracy: 49.25 -> 70.89 | Improved |
| add LeadNet instead of CSL on digit attribute | CSL · Colored MNIST digit attribute | accuracy: 61.60 -> 70.89 | Improved |
| use LeadNet without attribute labels instead of PAN-Supervised with attribute labels | PAN-Supervised · UT-Zappos50K 5-shot 5-way | accuracy: 46.92 -> 48.29 | Improved |
| add attribute supervision to ProtoNet (ProtoAttribute) | ProtoNet · UT-Zappos50K 1-shot 5-way | accuracy: 38.19 -> 38.75 | Mixed |
| add LeadNet instead of FEAT on MiniImageNet | FEAT · MiniImageNet 1-shot 5-way | accuracy: 66.78 -> 67.32 | Improved |
| add LeadNet instead of CGCS-Trans on CUB | CGCS-Trans · CUB 1-shot 5-way | accuracy: 76.69 -> 79.05 | Improved |
| add LeadNet instead of wave-SAN with ResNet-10 cross-domain 5-shot | wave-SAN · MiniImageNet to CUB cross-domain 5-shot 5-way (ResNet-10) | accuracy: 71.16 -> 70.88 | Hurt |
| add LeadNet instead of HSML on regression | HSML · Random-component regression 10-shot | MSE: 0.161 -> 0.121 | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- DeepEMD — DeepEMD: Few-Shot Image Classification With Differentiable Earth Mover's Distance and Structured Classifiers (CVPR 2020) · better on 6 of 6 shared results
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- A Dual Attention Network with Semantic Embedding for Few-Shot Learning (AAAI 2019 · similarity 0.87)
- Context-Aware Meta-Learning (ICLR 2024 · similarity 0.87)
- Unsupervised Meta-Learning via In-Context Learning (ICLR 2025 · similarity 0.87)
- Meta Networks (ICML 2017 · similarity 0.87)
- Unsupervised Learning via Meta-Learning (ICLR 2019 · similarity 0.86)
- Revisiting Unsupervised Meta-Learning via the Characteristics of Few-Shot Tasks (TPAMI 2022 · 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.
