Adapting Stochastic Search For Real-time Dynamic Weighted Constraint Satisfaction

TL;DR

This paper introduces two new algorithms, DMaxWalkSat and RDMaxWalkSat, designed for dynamic, weighted constraint satisfaction problems, with the latter supporting real-time, anytime solutions, improving solution quality and runtime efficiency.

Contribution

The paper presents novel algorithms tailored for dynamic, weighted constraint satisfaction, including an anytime version for real-time applications, advancing existing solver capabilities.

Findings

DMaxWalkSat outperforms MaxWalkSat in solution quality and runtime.

RDMaxWalkSat supports anytime, real-time constraint satisfaction.

Algorithms enhance tools for problems modeled as constraint satisfaction.

Abstract

This work presents two new algorithms for performing constraint satisfaction. The first algorithm presented, DMaxWalkSat, is a constraint solver specialized for solving dynamic, weighted constraint satisfaction problems. The second algorithm, RDMaxWalkSat, is a derivative of DMaxWalkSat that has been modified into an anytime algorithm, and hence support realtime constraint satisfaction. DMaxWalkSat is shown to offer performance advantages in terms of solution quality and runtime over its parent constraint solver, MaxWalkSat. RDMaxWalkSat is shown to support anytime operation. The introduction of these algorithms brings another tool to the areas of computer science that naturally represent problems as constraint satisfaction problems, an example of which is the robust coherence algorithm.