在高铁上刷NeurIPS paper,很多都很有意思,陆续推荐。
比如,AI for Science 需要在回答问题之前提出问题,需要在确定性之下寻找惊喜。
推荐一篇论文 Agarwal, Dhruv, Bodhisattwa Prasad Majumder, Reece Adamson, et al. "AutoDiscovery: Open-Ended Scientific Discovery via Bayesian Surprise." arXiv:2507.00310. Preprint, arXiv, February 12, 2026.
> The promise of autonomous scientific discovery (ASD) hinges not only on answering questions, but also on knowing which questions to ask.
> A key challenge, then, is that of search—which hypotheses should be investigated next that will lead to novel, impactful scientific discoveries?
> In this work, we address these questions and propose AUTODISCOVERY—a method for open-ended ASD that is guided by Bayesian surprise, which quantifies how data affects a natural or artificial observer by measuring the distance between its posterior and prior belief distributions