Research map / Label-efficient and robust learning / Test-time adaptation
M-L2O: Towards Generalizable Learning-to-Optimize by Test-Time Fast Self-Adaptation
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 meta-adaptation to learn optimizer initialization, then adapt for 5 steps | L2O optimizer · LASSO, σ=50 | minimum logarithm loss: -3.285 vs -3.181 | Helped |
| Use meta-adaptation to learn optimizer initialization, then adapt for 5 steps | L2O optimizer · LASSO, σ=100 | minimum logarithm loss: -2.408 vs -1.961 | Helped |
| Use meta-adaptation to learn optimizer initialization, then adapt for 5 steps | L2O optimizer · LASSO, σ=200 | minimum logarithm loss: -1.396 vs -0.982 | Helped |
| Use meta-adaptation to learn optimizer initialization, then adapt for 5 steps | L2O optimizer · LASSO, σ=10 | minimum logarithm loss: -3.712 vs -4.077 | Hurt |
| Use meta-adaptation to learn optimizer initialization, then adapt for 5 steps | L2O optimizer · LASSO, σ=25 | minimum logarithm loss: -3.433 vs -4.019 | Hurt |
| Adapt vanilla L2O on the adaptation tasks for 5 steps instead of using meta-trained initialization | L2O optimizer · LASSO out-of-distribution tasks | minimum logarithm loss | Hurt |
| Skip adaptation after training on training tasks | Direct Transfer · LASSO out-of-distribution tasks | minimum logarithm loss | Hurt |
| Adapt on tasks similar to the training distribution | M-L2O · LASSO test distribution with σ=100 | testing performance | Helped |
| Increase deviation of test tasks from training tasks | M-L2O · LASSO, σ greater than 50 | logarithmic loss: exceed 0.1 | Helped |
| Adapt on training-like tasks rather than only testing-like tasks | M-L2O · LASSO | testing performance | Helped |
Closest papers by meaning
- Training Stronger Baselines for Learning to Optimize (NeurIPS 2020 · similarity 0.93)
- Learning Provably Improves the Convergence of Gradient Descent (NeurIPS 2025 · similarity 0.87)
- Delve into the Applicability of Advanced Optimizers for Multi-Task Learning (NeurIPS 2026 · similarity 0.87)
- Towards Robust Learning to Optimize with Theoretical Guarantees (CVPR 2024 · similarity 0.86)
- MetaBox-v2: A Unified Benchmark Platform for Meta-Black-Box Optimization (NeurIPS 2025 · similarity 0.86)
- MetaOptimize: A Framework for Optimizing Step Sizes and Other Meta-parameters (ICML 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.