TacticToe: Learning to Prove with Tactics
Gauthier, Thibault, Kaliszyk, Cezary, Urban, Josef, Kumar, Ramana, Norrish, Michael
–arXiv.org Artificial Intelligence
Tactics analyze the current proof state (goal and assumptions) and apply non-trivial proof transformations. Formalized proofs take advantage of different levels of automation which are in increasing order of generality: specialized rules, theory-based strategies and general purpose strategies. Thanks to progress in proof automation, developers can delegate more and more complicated proof obligations to general purpose strategies. Those are implemented by automated theorem provers (ATPs) such as E prover [32]. Communication between an ITP and ATPs is made possible by a "hammer" system [4,14]. It acts as an interface by performing premise selection, translation and proof reconstruction. Yet, ATPs are not flawless and more precise user-guidance, achieved by applying a particular sequence of specialized rules, is almost always necessary to develop a mathematical theory.
arXiv.org Artificial Intelligence
Dec-1-2021
- Country:
- Europe > Germany (0.46)
- North America > United States (0.46)
- Genre:
- Research Report (0.64)
- Technology: