Training a First-Order Theorem Prover from Synthetic Data

A major challenge in applying machine learning to automated theorem proving is the scarcity of training data, which is a key ingredient in training successful deep learning models. To tackle this problem, we propose an approach that relies on training purely with synthetically generated theorems, without any human data aside from axioms. We use these theorems to train a neurally-guided saturationbased prover. Our neural prover outperforms the state-of-the-art E-prover on this synthetic data in both time and search steps, and shows significant transfer to the unseen human-written theorems from the TPTP library, where it solves 72% of first-order problems without equality.

[1]  Melvin Fitting,et al.  First-Order Logic and Automated Theorem Proving , 1990, Graduate Texts in Computer Science.

[2]  Ilya Sutskever,et al.  Generative Language Modeling for Automated Theorem Proving , 2020, ArXiv.

[3]  Cezary Kaliszyk,et al.  Deep Network Guided Proof Search , 2017, LPAR.

[4]  Simon Cruanes,et al.  Faster, Higher, Stronger: E 2.3 , 2019, CADE.

[5]  Demis Hassabis,et al.  A general reinforcement learning algorithm that masters chess, shogi, and Go through self-play , 2018, Science.

[6]  Mark Chen,et al.  Language Models are Few-Shot Learners , 2020, NeurIPS.

[7]  Andrew K. Lampinen,et al.  Automated curricula through setter-solver interactions , 2019, ArXiv.

[8]  Sarah M. Loos,et al.  HOList: An Environment for Machine Learning of Higher-Order Theorem Proving (extended version) , 2019, ArXiv.

[9]  Lukasz Kaiser,et al.  Attention is All you Need , 2017, NIPS.

[10]  Alec Radford,et al.  Scaling Laws for Neural Language Models , 2020, ArXiv.

[11]  Geoff Sutcliffe.  The TPTP Problem Library and Associated Infrastructure , 2017, Journal of Automated Reasoning.

[12]  Jason Rute,et al.  Proof Artifact Co-training for Theorem Proving with Language Models , 2021, ArXiv.

[13]  Sergey Levine,et al.  Emergent Complexity and Zero-shot Transfer via Unsupervised Environment Design , 2020, NeurIPS.

[14]  Ilya Kostrikov,et al.  Intrinsic Motivation and Automatic Curricula via Asymmetric Self-Play , 2017, ICLR.

[15]  Jia Deng,et al.  Learning to Prove Theorems via Interacting with Proof Assistants , 2019, ICML.

[16]  Jia Deng,et al.  Learning to Prove Theorems by Learning to Generate Theorems , 2020, NeurIPS.

[17]  Jürgen Schmidhuber,et al.  Simple algorithmic theory of subjective beauty, novelty, surprise, interestingness, attention, curiosity, creativity, art, science, music, jokes (特集 高次機能の学習と創発--脳・ロボット・人間研究における新たな展開) , 2009 .

[18]  Pierre-Yves Oudeyer,et al.  Intrinsically Motivated Goal Exploration Processes with Automatic Curriculum Learning , 2017, J. Mach. Learn. Res..

[19]  Andrei Voronkov,et al.  First-Order Theorem Proving and Vampire , 2013, CAV.

[20]  Xavier Bresson,et al.  A Generalization of Transformer Networks to Graphs , 2020, ArXiv.

[21]  Gordon Plotkin,et al.  A Note on Inductive Generalization , 2008 .

[22]  Larry Wos,et al.  Otter - The CADE-13 Competition Incarnations , 1997, Journal of Automated Reasoning.

[23]  Charisma Lee.  A completeness theorem and a computer program for finding theorems derivable from given axioms , 1967 .

[24]  Andrei Voronkov,et al.  The design and implementation of VAMPIRE , 2002, AI Commun..

[25]  Geoffrey E. Hinton,et al.  Rectified Linear Units Improve Restricted Boltzmann Machines , 2010, ICML.