Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Surrogate Assisted Monte Carlo Tree Search in Combinatorial Optimization

Saeid Amiri, Parisa Zehtabi, Danial Dervovic, Michael Cashmore

TL;DR

Results suggest that MCTS supported by a fast surrogate function can generate solutions faster while maintaining a consistent solution compared to MCTS that does not benefit from the surrogate function.

Abstract

Industries frequently adjust their facilities network by opening new branches in promising areas and closing branches in areas where they expect low profits. In this paper, we examine a particular class of facility location problems. Our objective is to minimize the loss of sales resulting from the removal of several retail stores. However, estimating sales accurately is expensive and time-consuming. To overcome this challenge, we leverage Monte Carlo Tree Search (MCTS) assisted by a surrogate model that computes evaluations faster. Results suggest that MCTS supported by a fast surrogate function can generate solutions faster while maintaining a consistent solution compared to MCTS that does not benefit from the surrogate function.

Surrogate Assisted Monte Carlo Tree Search in Combinatorial Optimization

TL;DR

Results suggest that MCTS supported by a fast surrogate function can generate solutions faster while maintaining a consistent solution compared to MCTS that does not benefit from the surrogate function.

Abstract

Industries frequently adjust their facilities network by opening new branches in promising areas and closing branches in areas where they expect low profits. In this paper, we examine a particular class of facility location problems. Our objective is to minimize the loss of sales resulting from the removal of several retail stores. However, estimating sales accurately is expensive and time-consuming. To overcome this challenge, we leverage Monte Carlo Tree Search (MCTS) assisted by a surrogate model that computes evaluations faster. Results suggest that MCTS supported by a fast surrogate function can generate solutions faster while maintaining a consistent solution compared to MCTS that does not benefit from the surrogate function.
Paper Structure (12 sections, 3 equations, 4 figures, 2 algorithms)

This paper contains 12 sections, 3 equations, 4 figures, 2 algorithms.

Table of Contents

  1. Introduction
  2. Related Work
  3. Surrogate-assisted optimization
  4. Problem Statement
  5. Framework
  6. Node representation:
  7. Algorithm:
  8. Experiments
  9. Dataset:
  10. Evaluation Functions:
  11. Conclusion & Future Work
  12. Disclaimer.

Figures (4)

  • Figure 1: Surrogate assisted Monte Carlo Tree Search (SMCTS) where an occasional reevaluation step refines the node values.
  • Figure 2: SMCTS where an occasional reevaluation step refines the node values. The horizontal axis represents the number of stores that need to be removed.
  • Figure 3: SMCTS with various surrogate errors. The vertical axis is the ratio of surrogate function $F_s$ evaluations to the total evaluations. The horizontal axis represent surrogate functions with increasing normalized RMSEs.
  • Figure 4: Evaluation of the consistency of stores selected by SMCTS vs. MCTS. The vertical axis shows the number of the selected stores by SMCTS being different from MCTS. The results are the average of $10$ counties that are randomly selected.