Neural Networks for Proof-Pattern Recognition

* Final gross prices may vary according to local VAT.

Get Access

Abstract

We propose a new method of feature extraction that allows to apply pattern-recognition abilities of neural networks to data-mine automated proofs. We propose a new algorithm to represent proofs for first-order logic programs as feature vectors; and present its implementation. We test the method on a number of problems and implementation scenarios, using three-layer neural nets with backpropagation learning.