Research map / Label-efficient and robust learning / Knowledge distillation
Meta-Learning with Self-Improving Momentum Target
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 momentum network as target (no distillation) (ablation) | MAML · mini-ImageNet 5-shot Conv4 | accuracy (%): 63.27 -> 66.12 | Helped |
| add distillation loss to momentum network (ablation) | MAML + momentum · mini-ImageNet 5-shot Conv4 | accuracy (%): 66.12 -> 66.52 | No clear effect |
| add dropout on top of momentum + distillation (ablation) | SiMT · mini-ImageNet 5-shot Conv4 | accuracy (%): 66.52 -> 68.74 | Helped |
| use dropout with momentum but without distillation (ablation) | MAML + momentum · mini-ImageNet 5-shot Conv4 | accuracy (%): 66.12 -> 65.25 | Hurt |
| add SiMT | MAML · mini-ImageNet 5-shot Conv4 | accuracy (%): 63.27 -> 68.74 | Helped |
| add SiMT | MAML · ShapeNet 10-shot regression | angular error: 29.555 -> 18.913 | Helped |
| add SiMT | MAML · Pascal 10-shot regression | MSE: 2.612 -> 1.462 | Helped |
| add SiMT | ProtoNet ResNet-12 · tiered-ImageNet 5-shot | accuracy (%): 79.94 -> 81.82 | Helped |
| add SiMT | MetaSGD ResNet-12 · mini-ImageNet->Cars 1-shot | accuracy (%): 34.52 -> 33.92 | Hurt |
| add SiMT | MAML ResNet-12 · mini-ImageNet->CUB 5-shot | accuracy (%): 56.17 -> 59.22 | Helped |
| replace Bootstrap target with momentum target | MAML Conv4 · mini-ImageNet 1-shot | accuracy (%): 48.68 -> 51.49 | Helped |
| replace Bootstrap target with momentum target | MAML Conv4 · tiered-ImageNet 1-shot | accuracy (%): 49.34 -> 52.51 | Helped |
Closest papers by meaning
- Towards Enabling Meta-Learning from Target Models (NeurIPS 2021 · similarity 0.88)
- Large-Scale Meta-Learning with Continual Trajectory Shifting (ICML 2021 · similarity 0.87)
- Meta-learning with an Adaptive Task Scheduler (NeurIPS 2021 · similarity 0.86)
- MetaOptimize: A Framework for Optimizing Step Sizes and Other Meta-parameters (ICML 2025 · similarity 0.84)
- Provable Meta-Learning with Low-Rank Adaptations (NeurIPS 2025 · similarity 0.84)
- Memory-Reduced Meta-Learning with Guaranteed Convergence (AAAI 2025 · similarity 0.84)
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.