In plain words: A tree-shaped neural network built inside the HOL4 proof assistant reads formulas as trees, making it a natural fit for learning functions over formulas. Its speed and accuracy were compared with other machine-learning predictors on arithmetic expression evaluation and judging propositional formulas true or false.

Abstract · Tree Neural Networks in HOL4

We present an implementation of tree neural networks within the proof assistant HOL4. Their architecture makes them naturally suited for approximating functions whose domain is a set of formulas. We measure the performance of our implementation and compare it with other machine learning predictors on the tasks of evaluating arithmetical expressions and estimating the truth of propositional formulas.

Thibault Gauthier
arXiv:2009.01827 · cs.NE · submitted Sep 3, 2020
abstract · pdf · html