arXiv：cs.AI· SiYuan Ma, Albert Gao, Chunzheng Zhu, Xin Yan, Wenlong Zhang, Wenxin Zhang, Luqi Gong, Tianlin Li, Qixin Zhang·· 6 小时前AI 评分43
让理论学会“成长”：超越初始假设空间的实验发现
Learning to Outgrow a Theory: Experimental Discovery Beyond the Initial Hypothesis Space
AI 导读
研究提出“实验模型类修订”方法，让发现策略同时提出结构编辑和检验该编辑是否必要的诊断实验，并用随时有效的序列证据在现有假设类被拒绝后触发结构修订。在 400 个受控动力学环境中，该联合策略以 32 次真实实验达到 89.5% 精确恢复率，比最强匹配基线提升 10.0 个百分点。当真实机制超出编辑语法时，该方法有 88% 的概率检测出库不足，误支持率仅 5.5%。
正文
Abstract:Scientific discovery systems typically optimize experiments within a fixed hypothesis space. This creates a failure mode when all available candidates omit the same missing mechanism: candidate disagreement can collapse even while the model class is systematically wrong. We formulate experimental model-class revision, in which a discovery policy jointly proposes a structural edit and a diagnostic experiment that tests whether that edit is necessary. The method couples a class-level distinguishability objective, in which one shared parameterization must explain all selected experiments, with anytime-valid sequential evidence that triggers structural revision only after the current class is rejected. On 400 held-out controlled dynamical environments, the joint policy reaches 89.5% exact recovery with a budget of 32 real experiments, improving the strongest matched baseline by 10.0 percentage points while requiring fewer executed experiments and candidate fits. The learned revision-experiment pairing transfers across unseen mechanism combinations, held-out but expressible primitives, parameter extrapolation, and shifted experiment costs; when the true mechanism is outside the edit grammar, it detects library insufficiency in 88% of cases with a 5.5% false-support rate. Revision gains also transfer to ODEBench and ODEBase model-library tasks, as well as DiscoverPhysics worlds. These results support a view of scientific discovery in which deciding what mechanisms a theory should make expressible and where to collect evidence are treated as a single sequential decision problem.
| Subjects: | Artificial Intelligence (cs.AI) |
| Cite as: | arXiv:2610.07627 [cs.AI] |
| (or arXiv:2610.07627v1 [cs.AI] for this version) | |
| https://doi.org/10.48550/arXiv.2610.07627 arXiv-issued DOI via DataCite (pending registration) |
Submission history
From: SiYuan Ma [view email]
[v1]
Tue, 6 Oct 2026 02:19:53 UTC (7,567 KB)
来源：arXiv：cs.AI · arxiv.org