Download PDFOpen PDF in browser

Breeding Theorem Proving Heuristics with Genetic Algorithms

12 pagesPublished: December 18, 2015

Abstract

First-order theorem provers have to search for proofs in an infinite
space of possible derivations. Proof search heuristics play a vital
role for the practical performance of these systems. In the current
generation of saturation-based theorem provers like SPASS, E,
Vampire or Prover~9, one of the most important decisions is the
selection of the next clause to process with the given clause
algorithms. Provers offer a wide variety of basic clause evaluation
functions, which can often be parameterized and combined in many
different ways. Finding good strategies is usually left to the users
or developers, often backed by large-scale experimental
evaluations. We describe a way to automatize this process using
genetic algorithms, evaluating a population of different strategies
on a test set, and applying mutation and crossover operators to good
strategies to create the next generation. We describe the design and
experimental set-up, and report on first promising results.

Keyphrases: automated theorem proving, genetic algorithms, heuristic search

In: Georg Gottlob, Geoff Sutcliffe and Andrei Voronkov (editors). GCAI 2015. Global Conference on Artificial Intelligence, vol 36, pages 263-274.

BibTeX entry
@inproceedings{GCAI2015:Breeding_Theorem_Proving_Heuristics,
  author    = {Simon Schäfer and Stephan Schulz},
  title     = {Breeding Theorem Proving Heuristics with Genetic Algorithms},
  booktitle = {GCAI 2015. Global Conference on Artificial Intelligence},
  editor    = {Georg Gottlob and Geoff Sutcliffe and Andrei Voronkov},
  series    = {EPiC Series in Computing},
  volume    = {36},
  publisher = {EasyChair},
  bibsource = {EasyChair, https://easychair.org},
  issn      = {2398-7340},
  url       = {/publications/paper/M5},
  doi       = {10.29007/gms9},
  pages     = {263-274},
  year      = {2015}}
Download PDFOpen PDF in browser