Online machine learning techniques for Coq: a comparison

DOI10.1007/978-3-030-81097-9_5zbMATH Open1485.68288arXiv2104.05207OpenAlexW3186870373MaRDI QIDQ2128797FDOQ2128797

Authors: Liao Zhang, Lasse Blaauwbroek, Bartosz Piotrowski, Prokop Černỳ, Cezary Kaliszyk, Josef Urban 

Publication date: 22 April 2022

Abstract: We present a comparison of several online machine learning techniques for tactical learning and proving in the Coq proof assistant. This work builds on top of Tactician, a plugin for Coq that learns from proofs written by the user to synthesize new proofs. Learning happens in an online manner, meaning that Tactician's machine learning model is updated immediately every time the user performs a step in an interactive proof. This has important advantages compared to the more studied offline learning systems: (1) it provides the user with a seamless, interactive experience with Tactician and, (2) it takes advantage of locality of proof similarity, which means that proofs similar to the current proof are likely to be found close by. We implement two online methods, namely approximate k-nearest neighbors based on locality sensitive hashing forests and random decision forests. Additionally, we conduct experiments with gradient boosted trees in an offline setting using XGBoost. We compare the relative performance of Tactician using these three learning methods on Coq's standard library.

Full work available at URL: https://arxiv.org/abs/2104.05207

Recommendations

• The Tactician. A seamless, interactive tactic learner and prover for Coq
• TacticToe: learning to reason with HOL4 tactics
• scientific article
• MaSh: machine learning for Sledgehammer
• Machine learning for first-order theorem proving

zbMATH Keywords

machine learningrandom forestonline learningCoqinteractive theorem provinggradient boosted trees

Mathematics Subject Classification ID

Learning and adaptive systems in artificial intelligence (68T05) Online algorithms; streaming algorithms (68W27) Theorem proving (automated and interactive theorem provers, deduction, resolution, etc.) (68V15)

Cites Work

• ATPboost: learning premise selection in binary setting with ATP feedback
• TacticToe: learning to reason with HOL4 tactics
• Random forests
• A proof strategy language and proof script generation for Isabelle/HOL
• Making data structures persistent
• Multidimensional binary search trees used for associative searching
• Approximate nearest neighbor: towards removing the curse of dimensionality
• The Tactician. A seamless, interactive tactic learner and prover for Coq
• ENIGMA-NG: efficient neural and gradient-boosted inference guidance for \(\mathrm{E}\)
• Random forests for premise selection
• TacticToe: learning to prove with tactics

Cited In (3)

• Title not available (Why is that?)
• Learning Proof Transformations and Its Applications in Interactive Theorem Proving
• Machine-learned premise selection for Lean

Uses Software

• Coq
• Tactician
• XGBoost
• ATPboost
• TacticToe
• ENIGMA
• HOList

This page was built for publication: Online machine learning techniques for Coq: a comparison

Report a bug (only for logged in users!)Click here to report a bug for this page (MaRDI item Q2128797)