Breaking Multi-Task Curse: Reward-Weighted Evolution for Black-Box Many-Task Optimization

Email: int_lyc@cug.edu.cn

Reference

@InProceedings{Li2026MES-RET,
  author    = {Li, Yanchi and Liu, Jiao and Gong, Wenyin and Gu, Qiong and Zhao, Yue and Ong, Yew-Soon},
  booktitle = {Forty-third International Conference on Machine Learning},
  title     = {Breaking Multi-Task Curse: Reward-Weighted Evolution for Black-Box Many-Task Optimization},
  year      = {2026},
}

Note

The MTO-Platform (MToP) has incorporated the relevant algorithm and problems at:

MES-RET Algorithm

Many-task CEC Benchmark

Policy Search Task Gymnasium

MToP provides convenient utility functions. We highly recommend using it for code running. https://github.com/intLyc/MTO-Platform

Rendered GIFs click here.

Reward: 99

Reward: -117

Reward: 500

Reward: -77

Reward: 273

Reward: 299

Reward: 9360

Reward: 6393

Reward: 3541

Reward: 4671

Reward: 364

Reward: 2849

Reward: 551

Reward: 191258