Research map / Optimization / Optimizers
Meta-Learning With a Geometry-Adaptive Preconditioner
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 |
|---|---|---|---|
| use GAP instead of the strongest reported baseline, PAMELA | few-shot sinusoid regression · Sinusoid regression, 5-shot | MSE: 0.54±0.06 -> 0.16±0.04 | Helped |
| use GAP instead of the strongest reported baseline, PAMELA | few-shot sinusoid regression · Sinusoid regression, 10-shot | MSE: 0.41±0.04 -> 0.04±0.01 | Helped |
| use GAP instead of the strongest reported baseline, PAMELA | few-shot sinusoid regression · Sinusoid regression, 20-shot | MSE: 0.17±0.03 -> 0.01±0.01 | Helped |
| use GAP instead of PAMELA | mini-ImageNet classification · 5-way 1-shot | accuracy: 53.50 ± 0.89 -> 54.86 ± 0.85 | Helped |
| use GAP instead of MeTAL | mini-ImageNet classification · 5-way 5-shot | accuracy: 70.52 ± 0.29 -> 71.55±0.61 | Helped |
| use GAP instead of WarpGrad | tiered-ImageNet classification · 5-way 1-shot | accuracy: 57.20 ± 0.90 -> 57.60 ± 0.93 | Helped |
| use GAP instead of PAMELA | tiered-ImageNet classification · 5-way 5-shot | accuracy: 74.39 ± 0.71 -> 74.90 ± 0.68 | Helped |
| replace full adaptive GAP with Approximate GAP (ablation) | few-shot classification · mini-ImageNet and tiered-ImageNet | accuracy: 54.86 ± 0.85 -> 53.52 ± 0.88; 71.55±0.61 -> 70.75 ± 0.67; 57.60 ± 0.93 -> 56.86 ± 0.91; 74.90 ± 0.68 -> 74.41 ± 0.72 | Hurt |
| replace full adaptive GAP with Approximate GAP (ablation) | cross-domain few-shot classification · tiered-ImageNet, CUB, and Cars | accuracy: 58.56±0.93 -> 57.47 ± 0.99; 72.82±0.77 -> 71.66 ± 0.76; 44.74±0.75 -> 43.77 ± 0.79; 64.88±0.72 -> 62.92 ± 0.73; 38.44±0. | Hurt |
| do not apply the learned GAP preconditioner (ablation) | GAP-trained mini-ImageNet model · 5-way 1-shot | accuracy: 54.86 ± 0.85 -> 48.23 ± 0.80 | Hurt |
| do not apply the learned GAP preconditioner (ablation) | GAP-trained mini-ImageNet model · 5-way 5-shot | accuracy: 71.55 ± 0.61 -> 65.80 ± 0.75 | Hurt |
| use mode-1 unfolding rather than mode-2, mode-3, or mode-4 (ablation) | GAP gradient matrix construction · mini-ImageNet, 5-way 1-shot | accuracy: 53.02 ± 0.87; 51.23 ± 0.76; 51.45 ± 0.77 -> 54.86 ± 0.85 | Helped |
Closest papers by meaning
- Convergence of Meta-Learning with Task-Specific Adaptation over Partial Parameters (NeurIPS 2020 · similarity 0.88)
- Riemannian Geometric-based Meta Learning (AAAI 2025 · similarity 0.88)
- MetaDiff: Meta-Learning with Conditional Diffusion for Few-Shot Learning (AAAI 2024 · similarity 0.87)
- Riemannian Preconditioned LoRA for Fine-Tuning Foundation Models (ICML 2024 · similarity 0.86)
- Enabling Few-Shot Learning with PID Control: A Layer Adaptive Optimizer (ICML 2024 · similarity 0.86)
- Memory-Reduced Meta-Learning with Guaranteed Convergence (AAAI 2025 · similarity 0.85)
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.