Research map / Label-efficient and robust learning / Zero-/few-shot and meta-learning
Instance-wise Adaptive Scheduling via Derivative-Free Meta-Learning
Hefang Qing, Miao Zhang, Yaoxin Wu, Weinan Huang, Jianhao Yang, Wen Song, Gang Wang
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 derivative-free meta-learning fine-tuning versus EAS | DANIEL · FJSP SD1 10x5 greedy | Optimality gap (%): 7.68% -> 7.06% | Improved |
| add derivative-free meta-learning fine-tuning versus EAS | DANIEL · FJSP SD2 10x5 sampling | Optimality gap (%): 8.80% -> 6.59% | Improved |
| add derivative-free meta-learning fine-tuning versus EAS | DANIEL · FJSP SD2 40x10 unseen sampling | Optimality gap (%): -3.72% -> -5.20% | Improved |
| add meta-learning on top of derivative-free ES training (ablation) | DANIEL · FJSP SD2 10x5 | Optimality gap (%): 14.38% -> 13.18% | Improved |
| add meta-learning on top of derivative-free ES training (ablation) | DANIEL · FJSP SD2 20x10 | Optimality gap (%): 24.20% -> 21.22% | Improved |
| add derivative-free ES training versus gradient-based AS/EAS (ablation) | DANIEL · FJSP SD2 10x5 | Optimality gap (%): 16.65% -> 14.38% | Improved |
| use derivative-free meta-learning instead of FOMAML (greedy) (ablation) | DANIEL · FJSP SD2 10x5 greedy | Optimality gap (%): 20.91% -> 13.18% | Improved |
| use derivative-free meta-learning instead of FOMAML (sampling) (ablation) | DANIEL · FJSP SD2 40x10 sampling | Optimality gap (%): -1.02% -> -5.20% | Improved |
| use GPU-parallel ES instead of Ray CPU multithreading (ablation) | ES fitness evaluation · FJSP 20x10 | forward pass time for 100 individuals: 122.3s -> 3.8s | Improved |
| add MC averaging and MC best-sample inner-loop estimators over single-sample NES (ablation) | DFO inner loop · FJSP SD2 10x5 sampling | Optimality gap (%) | Improved |
| add derivative-free meta-learning on public benchmark (greedy, SD1 10x5 model) | DANIEL · mk benchmark | Optimality gap (%): 6.79% -> 4.83% | Improved |
Closest papers by meaning
- Meta-SAGE: Scale Meta-Learning Scheduled Adaptation with Guided Exploration for Mitigating Scale Shift on Combinatorial Optimization (ICML 2023 · similarity 0.88)
- Towards Learned Optimization Free Lunch (ICLR 2026 · similarity 0.86)
- Meta-Reinforcement Learning with Universal Policy Adaptation: Provable Near-Optimality under All-task Optimum Comparator (NeurIPS 2024 · similarity 0.86)
- MetaBox-v2: A Unified Benchmark Platform for Meta-Black-Box Optimization (NeurIPS 2025 · similarity 0.85)
- Meta-Reinforcement Learning with Adaptation from Human Feedback via Preference-Order-Preserving Task Embedding (ICML 2025 · similarity 0.85)
- Where Does Warm-Up Come From? Adaptive Scheduling for Norm-Constrained Optimizers (NeurIPS 2026 · 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.
