First Experiments with Neural cvc5

TL;DR

This paper introduces a neural network-guided approach to enhance the cvc5 solver's ability to handle first-order problems, demonstrating improved performance through a graph neural network that efficiently guides instantiation choices.

Contribution

It develops a fast, CPU-compatible graph neural network to guide quantifier instantiation in cvc5, trained on proof data to improve solver efficiency.

Findings

Neural guidance improves cvc5's problem-solving performance.

The graph neural network runs efficiently on CPU.

Training on proof data enhances instantiation effectiveness.

Abstract

he cvc5 solver is today one of the strongest systems for solving first order problems with theories but also without them. In this work we equip its enumeration-based instantiation with a neural network that guides the choice of the quantified formulas and their instances. For that we develop a relatively fast graph neural network that repeatedly scores all available instantiation options with respect to the available formulas. The network runs directly on a CPU without the need for any special hardware. We train the neural guidance on a large set of proofs generated by the e-matching instantiation strategy and evaluate its performance on a set of previously unseen problems.