Reinforcement Learning of Theorem Proving

被引:0
|
作者
Kaliszyk, Cezary [1 ]
Urban, Josef [2 ]
Michalewski, Henryk [3 ]
Olsak, Mirek [4 ]
机构
[1] Univ Innsbruck, Innsbruck, Austria
[2] Czech Tech Univ, Prague, Czech Republic
[3] Univ Warsaw, Inst Math, Polish Acad Sci, Warsaw, Poland
[4] Charles Univ Prague, Prague, Czech Republic
基金
欧洲研究理事会;
关键词
GAME; GO;
D O I
暂无
中图分类号
TP18 [人工智能理论];
学科分类号
081104 ; 0812 ; 0835 ; 1405 ;
摘要
We introduce a theorem proving algorithm that uses practically no domain heuristics for guiding its connection-style proof search. Instead, it runs many Monte-Carlo simulations guided by reinforcement learning from previous proof attempts. We produce several versions of the prover, parameterized by different learning and guiding algorithms. The strongest version of the system is trained on a large corpus of mathematical problems and evaluated on previously unseen problems. The trained system solves within the same number of inferences over 40% more problems than a baseline prover, which is an unusually high improvement in this hard AI domain. To our knowledge this is the first time reinforcement learning has been convincingly applied to solving general mathematical problems on a large scale.
引用
收藏
页码:8836 / 8847
页数:12
相关论文
共 50 条
  • [21] Theorem proving for verification
    Harrison, John
    COMPUTER AIDED VERIFICATION, 2008, 5123 : 11 - 18
  • [22] Constraints and theorem proving
    Ganzinger, H
    Nieuwenhuis, R
    CONSTRAINTS IN COMPUTATIONAL LOGICS: THEORY AND APPLICATIONS, 2001, 2002 : 159 - 201
  • [23] Advances in theorem proving
    Kientzle, T
    DR DOBBS JOURNAL, 1997, 22 (03): : 16 - 16
  • [24] Theorem Proving Modulo
    Gilles Dowek
    Thérèse Hardin
    Claude Kirchner
    Journal of Automated Reasoning, 2003, 31 : 33 - 72
  • [25] Automated theorem proving
    Plaisted, David A.
    WILEY INTERDISCIPLINARY REVIEWS-COGNITIVE SCIENCE, 2014, 5 (02) : 115 - 128
  • [26] Unsound theorem proving
    Lynch, C
    COMPUTER SCIENCE LOGIC, PROCEEDINGS, 2004, 3210 : 473 - 487
  • [27] Machine Learning for First-Order Theorem Proving Learning to Select a Good Heuristic
    Bridge, James P.
    Holden, Sean B.
    Paulson, Lawrence C.
    JOURNAL OF AUTOMATED REASONING, 2014, 53 (02) : 141 - 172
  • [28] HOList: An Environment for Machine Learning of Higher-Order Theorem Proving
    Bansal, Kshitij
    Loos, Sarah
    Rabe, Markus
    Szegedy, Christian
    Wilcox, Stewart
    INTERNATIONAL CONFERENCE ON MACHINE LEARNING, VOL 97, 2019, 97
  • [29] Proving Termination by Dependency Pairs and Inductive Theorem Proving
    Carsten Fuhs
    Jürgen Giesl
    Michael Parting
    Peter Schneider-Kamp
    Stephan Swiderski
    Journal of Automated Reasoning, 2011, 47 : 133 - 160
  • [30] Proving Termination by Dependency Pairs and Inductive Theorem Proving
    Fuhs, Carsten
    Giesl, Juergen
    Parting, Michael
    Schneider-Kamp, Peter
    Swiderski, Stephan
    JOURNAL OF AUTOMATED REASONING, 2011, 47 (02) : 133 - 160