军事知识服务平台
Automated Conjecturing and Theorem Finding:A Survey
摘要 Automated conjecturing and theorem finding represents a fundamental challenge in artificial intelli-gence,involving the systematic generation of mathematical conjectures and theorems from given axioms and defi-nitions.This survey provides an overview of automated conjecturing and theorem finding methods.We catego-rize existing approaches into symbolic methods and data-driven methods.Symbolic methods,which include top-down lemma discovery and bottom-up theory exploration,leverage deductive reasoning and symbolic enumera-tion.Data-driven methods comprise statistical analogy learning and recent neural methods based on large lan-guage models,which learn patterns from large formal corpora,and also include efforts in scientific discovery from experimental data and theorem generation for synthetic data.We also discuss strategies and evaluation methods of interestingness.Finally,we identify current challenges and suggest some future research directions.This com-prehensive analysis aims to guide researchers in understanding the current state of automated conjecturing and identifying opportunities for advancing this critical area of mechanized mathematics and automated reasoning.
年,卷(期): 2026, 41(1)
机标分类号: TN925.93 TN915.04 G633
在线出版日期: 2026年6月29日