Applying Machine Learning to Heuristic Selection in an Automated Theorem Prover
Add to your list(s)
Download to your calendar using vCal
If you have a question about this talk, please contact Thomas Tuerk.
The speed of finding a proof, or whether a proof is found at all, by an automated theorem prover may be critically dependent on the heuristic used. A heuristic that works very well in one case may be slow or not work when seeking the proof of a different conjecture. Good heuristics are generally found by trial and error and their features are closely tied in with the detailed workings of the theorem prover. If automated theorem provers are to be useful to users beyond a small group of specialists the choice of heuristic must be automated.
The work to be described seeks to determine if there is a connection between structural features of the conjecture and axioms and the choice of the best heuristic to be used. Machine learning is used to find any such connection and embody it in software which may be used to automatically select a good heuristic.
This talk is part of the Computer Laboratory Automated Reasoning Group Lunches series.
This talk is included in these lists:
Note that exdirectory lists are not shown.
