DiscoveryRSI: Recursively Learning to Discover Valuable Hypotheses
Published in arXiv preprint, 2026
Recommended citation: Jianyuan Zhong, Chenwei Wu, Zhaochen Yu, Xixian Jiang, Jiajun Fan, Xuewei Li, Hao Xiao, Xilin Dai, Junyang Wang, Jiaru Zou, Zhiqing Cui, Boyuan Zheng, Jing Xiong, Zhen Zhang, Qiang Gao, Michael Morley, Luis Nakayama, Qiang Xu, Liyue Shen, Tao Gui, Leo Anthony Celi, Zhenfei Yin, Yingcheng Wu, Ling Yang. "DiscoveryRSI: Recursively Learning to Discover Valuable Hypotheses." arXiv preprint arXiv:2610.23839, 2026. https://arxiv.org/abs/2610.23839
DiscoveryRSI grows research hypotheses as trees through multi-agent evolutionary search, compiles the selected investigations into executable tasks, and improves recursively through hypothesis credit optimization, harness replay and rediscovery. It also introduces DFM-Bench, which scores hypothesis quality and cumulative research progress, and is evaluated across four life-science domains.