Scheduling Methods for Parallel Automated Theorem Proving

• Authors: Gernot Stenz;Andreas Wolf
• Venue: AI '00 Proceedings of the 13th Biennial Conference of the Canadian Society on Computational Studies of Intelligence: Advances in Artificial Intelligence
• Year: 2000

One of the key issues in automated theorem proving is the search for optimal proof strategies. Since there is not one uniform strategy which works optimally on all proof tasks, one is faced with the difficult problem of selecting a good strategy for a given task. Strategy parallelism is a way of circumventing this strategy selection problem. However, the problem of selecting the parallel strategies and distributing the available resources remains. Therefore we have developed a method for strategy evaluation and selection based on training data. We present a theorem prover system which has been automated with respect to the entire process of theorem prover application including automatic data generation, automatic schedule selection and classical automated theorem proving. In the theorem prover e-SETHEO, we present an implementation of such a system that, for the first time, can handle the necessary problem domain adaption fully automatically and which is an improvement of the prover which solved the largest number of problems in the MIX division of the CADE-16 ATP competition. This is followed by some experimental data produced with this system. We address the problem of test set extraction and give an assessment of our work as well as a lookout to future research issues.