Research map / Optimization / Optimizers
Meta-AdaM: An Meta-Learned Adaptive Optimizer with Momentum for Few-Shot Learning
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 heuristic momentum aggregation only (ablation) | MAML · Mini-ImageNet 5-way-1-shot | accuracy: 48.70±1.75 -> 48.28±0.49 | Hurt |
| add heuristic momentum aggregation only (ablation) | MAML · Mini-ImageNet 5-way-5-shot | accuracy: 63.11±0.92 -> 63.27±0.48 | No clear effect |
| add double look-ahead to momentum (ablation) | Meta-AdaM momentum · Mini-ImageNet 5-way-1-shot | accuracy: 48.28±0.49 -> 49.30±0.49 | Helped |
| add double look-ahead to momentum (ablation) | Meta-AdaM momentum · Mini-ImageNet 5-way-5-shot | accuracy: 63.27±0.48 -> 64.09±0.47 | Helped |
| add meta-learned adaptive learning rate only (ablation) | MAML · Mini-ImageNet 5-way-5-shot | accuracy: 63.11±0.92 -> 67.98±0.46 | Helped |
| add meta-learned adaptive learning rate only (ablation) | MAML · Mini-ImageNet 5-way-1-shot | accuracy: 48.70±1.75 -> 49.02±0.50 | No clear effect |
| combine momentum, look-ahead and adaptive lr (ablation) | MAML · Mini-ImageNet 5-way-1-shot | accuracy: 48.70±1.75 -> 51.64±0.49 | Helped |
| combine momentum, look-ahead and adaptive lr (ablation) | MAML · Mini-ImageNet 5-way-5-shot | accuracy: 63.11±0.92 -> 68.80±0.46 | Helped |
| replace LSTM learner with MLP using same inputs (ablation) | Meta-AdaM · Mini-ImageNet 5-way-1-shot | accuracy: 52.27± 0.49 -> 52.00 ± 0.49 | Mixed |
| replace LSTM learner with MLP using same inputs (ablation) | Meta-AdaM · Cifar100 5-way-5-shot | accuracy: 54.96 ± 0.49 -> 56.32 ± 0.49 | Helped |
| add dynamic class weighting (DW) (ablation) | Meta-AdaM · Mini-ImageNet 5-way-5-shot | accuracy: 68.80 ± 0.46 -> 70.70 ± 0.49 | Helped |
| add dynamic class weighting (DW) (ablation) | Meta-AdaM · Cifar100 5-way-5-shot | accuracy: 54.36 ± 0.50 -> 56.32 ± 0.49 | Helped |
Closest papers by meaning
- IO-Adam: Rethinking Memory-Efficient Adaptive Optimizers from Gradient Computation (ICML 2026 · similarity 0.90)
- MADA: Meta-Adaptive Optimizers Through Hyper-Gradient Descent (ICML 2024 · similarity 0.89)
- Dynamic Kernel Selection for Improved Generalization and Memory Efficiency in Meta-Learning (CVPR 2022 · similarity 0.88)
- SPAM: Spike-Aware Adam with Momentum Reset for Stable LLM Training (ICLR 2025 · similarity 0.88)
- A Comprehensive Framework for Analyzing the Convergence of Adam: Bridging the Gap with SGD (ICML 2025 · similarity 0.87)
- In Search of Adam’s Secret Sauce (NeurIPS 2025 · similarity 0.87)
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.