Research map / Reinforcement learning / Visual and representation RL
Breaking Multi-Task Curse: Reward-Weighted Evolution for Black-Box Many-Task Optimization
Yanchi Li, Jiao Liu, Wenyin Gong, Qiong Gu, Yue Zhao, Yew-Soon Ong
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 reward-weighted evaluation (vs uniform allocation) (ablation) | MES-RET · Synthetic optimization (87 tasks) | Friedman rank: 3.82 -> 3.13 | Improved |
| add reward-weighted transfer (ablation) | MES-RET · Synthetic optimization (87 tasks) | Friedman rank: 3.28 -> 3.13 | Mixed |
| add reward-weighted transfer (ablation) | MES-RET · Real-world engineering (42 tasks) | Friedman rank: 3.79 -> 3.54 | Improved |
| add reward-weighted transfer (ablation) | sep-MES-RET · Policy search (18 Gymnasium tasks) | Friedman rank: 5.08 -> 4.03 | Improved |
| add reward-weighted evaluation (ablation) | sep-MES-RET · Policy search (18 Gymnasium tasks) | Friedman rank: 5.42 -> 4.03 | Improved |
| naive multi/many-task methods vs single-task CMA-ES | SBCMAES, MTES-KG, AT-MFEA · Synthetic optimization (87 tasks) | Friedman rank: 3.90 (CMA-ES) vs 9.68 (SBCMAES), 4.97 (MTES-KG), 7.02 (AT-MFEA) | Hurt |
| add semantic parameter alignment transfer on heterogeneous architectures (ablation) | sep-MES-RET · HalfCheetah-v5 (MLP) | average reward: 4978.99 -> 5527.16 | Improved |
| add transfer across CNN+MLP architectures (ablation) | sep-MES-RET · Ant-v5 (CNN+MLP) | average reward: 1397.13 -> 1322.88 | Hurt |
| add transfer across CNN+MLP architectures (ablation) | sep-MES-RET · Walker2d-v5 (CNN+MLP) | average reward: 2716.71 -> 2845.42 | Improved |
Closest papers by meaning
- ARS: Adaptive Reward Scaling for Multi-Task Reinforcement Learning (ICML 2025 · similarity 0.87)
- Selective Task Group Updates for Multi-Task Optimization (ICLR 2025 · similarity 0.87)
- AMAGO-2: Breaking the Multi-Task Barrier in Meta-Reinforcement Learning with Transformers (NeurIPS 2024 · similarity 0.86)
- Towards Consistent Multi-Task Learning: Unlocking the Potential of Task-Specific Parameters (CVPR 2025 · similarity 0.86)
- R*: Efficient Reward Design via Reward Structure Evolution and Parameter Alignment Optimization with Large Language Models (ICML 2025 · similarity 0.86)
- Interpreting Neural Combinatorial Optimization via Evolving Programmatic Bottlenecks (NeurIPS 2026 · similarity 0.86)
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.
