Evolved preambles for MAX-SAT heuristics

TL;DR

This paper introduces evolved preambles, sequences of variable assignments, to improve MAX-SAT heuristics by providing better initial solutions, demonstrating significant performance gains through genetic algorithm optimization.

Contribution

It presents a novel method of evolving preambles to enhance MAX-SAT heuristics, outperforming traditional random initializations on benchmark instances.

Findings

Evolved preambles significantly improve MAX-SAT heuristic performance.

Genetic algorithms effectively optimize preambles for various heuristics.

Improved initial assignments lead to faster and more successful problem solving.

Abstract

MAX-SAT heuristics normally operate from random initial truth assignments to the variables. We consider the use of what we call preambles, which are sequences of variables with corresponding single-variable assignment actions intended to be used to determine a more suitable initial truth assignment for a given problem instance and a given heuristic. For a number of well established MAX-SAT heuristics and benchmark instances, we demonstrate that preambles can be evolved by a genetic algorithm such that the heuristics are outperformed in a significant fraction of the cases.