arXiv cs.AI· Kangjun Noh, Soyu Kim, Kyungwoo Song·· 4 小时前AI 评分31
CISE:用保形区间驱动自我进化,应对不完美代理奖励
Reliable Self-Evolution with Imperfect Proxy Rewards
AI 导读
针对高保真评估成本过高、低成本代理奖励易产生假阳性的问题,研究者提出保形区间驱动的自我进化方法 CISE,通过条件保形推断与逐迭代在线密度比估计为候选构建专属奖励区间。在材料科学的三个自我进化搜索任务中,CISE 返回的候选在高保真评估下全部为真阳性,而基线返回更多候选但包含假阳性。
正文 · 原文
Abstract:Large language model (LLM)-based self-evolving search is a promising approach to scientific discovery. However, high-fidelity evaluation of every candidate is prohibitively expensive in some domains. Self-evolving systems in such settings therefore rely on low-cost but imperfect proxy rewards, which may assign high scores to infeasible candidates. These false positives may contaminate both the final output and the feedback used to guide subsequent generations. This motivates statistically calibrated reward intervals for more reliable self-evolving search. We propose Conformal Interval-Driven Self-Evolution (CISE), which constructs candidate-specific reward intervals using conditional conformal inference and iteration-wise online density-ratio estimation. CISE uses conservative interval-based rewards for evolutionary feedback and returns candidates only when all required property intervals lie entirely within their respective feasible regions. We derive fixed-iteration coverage results under explicit assumptions of independence and covariate shift. We evaluate CISE on three self-evolving search tasks in materials science. In our experiments, all candidates returned by CISE are true positives under high-fidelity evaluation, whereas the baselines return more candidates but include false positives. These results highlight the value of a smaller, more precise shortlist when downstream validation budgets are limited. Our repository is available at this https URL.
| Comments: | Preprint |
| Subjects: | Artificial Intelligence (cs.AI) |
| Cite as: | arXiv:2610.02975 [cs.AI] |
| (or arXiv:2610.02975v2 [cs.AI] for this version) | |
| https://doi.org/10.48550/arXiv.2610.02975 arXiv-issued DOI via DataCite |
Submission history
From: Kangjun Noh [view email]
[v1]
Fri, 2 Oct 2026 08:07:27 UTC (3,516 KB)
[v2]
Tue, 17 Nov 2026 02:39:14 UTC (1,569 KB)
来源:arXiv cs.AI · arxiv.org