A Learning-Based Fact Selector for Isabelle/HOL

Jasmin Christian Blanchette (1, 2, 3) , David Greenaway (4, 5) , Cezary Kaliszyk (6) , Daniel Kühlwein (7) , Josef Urban (8)

1 VERIDIS - Modeling and Verification of Distributed Algorithms and Systems
2 MPII - Max-Planck-Institut für Informatik
3 MOSEL - Proof-oriented development of computer-based systems
4 NICTA - National ICT Australia [Sydney]
5 UNSW - University of New South Wales [Sydney]
6 Leopold Franzens Universität Innsbruck - University of Innsbruck
7 Radboud University [Nijmegen]
8 CTU - Czech Technical University in Prague

Sledgehammer integrates automatic theorem provers in the proof assistant Isa-belle/HOL. A key component, the fact selector, heuristically ranks the thousands of facts (lemmas, definitions, or axioms) available and selects a subset, based on syntactic similarity to the current proof goal. We introduce MaSh, an alternative that learns from successful proofs. New challenges arose from our "zero click" vision: MaSh integrates seamlessly with the users' workflow, so that they benefit from machine learning without having to install software, set up servers, or guide the learning. MaSh outperforms the old fact selector on large formalizations.

Domaines: Logique en informatique [cs.LO]

Jasmin Christian Blanchette, David Greenaway, Cezary Kaliszyk, Daniel Kühlwein, Josef Urban. A Learning-Based Fact Selector for Isabelle/HOL. Journal of Automated Reasoning, 2016, 57, pp.219 - 244. ⟨10.1007/s10817-016-9362-8⟩. ⟨hal-01386986⟩