Generalized Nested Rollout Policy Adaptation with Limited Repetitions

TL;DR

This paper introduces an improved Monte Carlo search algorithm, GNRPA, which limits repetitions of the best sequence to enhance exploration and performance across various combinatorial problems.

Contribution

The paper proposes a novel modification to GNRPA that prevents overly deterministic policies by limiting repetitions, leading to better results in multiple combinatorial tasks.

Findings

Improved performance on Inverse RNA Folding

Enhanced solutions for Traveling Salesman Problem with Time Windows

Better results on the Weak Schur problem

Abstract

Generalized Nested Rollout Policy Adaptation (GNRPA) is a Monte Carlo search algorithm for optimizing a sequence of choices. We propose to improve on GNRPA by avoiding too deterministic policies that find again and again the same sequence of choices. We do so by limiting the number of repetitions of the best sequence found at a given level. Experiments show that it improves the algorithm for three different combinatorial problems: Inverse RNA Folding, the Traveling Salesman Problem with Time Windows and the Weak Schur problem.