Graph-Supported Dynamic Algorithm Configuration for Multi-Objective Combinatorial Optimization

Fuente: arXiv
Saved in:
Bibliographic Details
Main Authors: Reijnen, Robbert, Wu, Yaoxin, Bukhsh, Zaharah, Zhang, Yingqian
Format: Preprint
Published: 2025
Subjects:
Online Access:
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866918033416519680
author Reijnen, Robbert
Wu, Yaoxin
Bukhsh, Zaharah
Zhang, Yingqian
author_facet Reijnen, Robbert
Wu, Yaoxin
Bukhsh, Zaharah
Zhang, Yingqian
contents Deep reinforcement learning (DRL) has been widely used for dynamic algorithm configuration, particularly in evolutionary computation, which benefits from the adaptive update of parameters during the algorithmic execution. However, applying DRL to algorithm configuration for multi-objective combinatorial optimization (MOCO) problems remains relatively unexplored. This paper presents a novel graph neural network (GNN) based DRL to configure multi-objective evolutionary algorithms. We model the dynamic algorithm configuration as a Markov decision process, representing the convergence of solutions in the objective space by a graph, with their embeddings learned by a GNN to enhance the state representation. Experiments on diverse MOCO challenges indicate that our method outperforms traditional and DRL-based algorithm configuration methods in terms of efficacy and adaptability. It also exhibits advantageous generalizability across objective types and problem sizes, and applicability to different evolutionary computation methods.
format Preprint
id arxiv_https___arxiv_org_abs_2505_16471
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle Graph-Supported Dynamic Algorithm Configuration for Multi-Objective Combinatorial Optimization
Reijnen, Robbert
Wu, Yaoxin
Bukhsh, Zaharah
Zhang, Yingqian
Neural and Evolutionary Computing
Machine Learning
Deep reinforcement learning (DRL) has been widely used for dynamic algorithm configuration, particularly in evolutionary computation, which benefits from the adaptive update of parameters during the algorithmic execution. However, applying DRL to algorithm configuration for multi-objective combinatorial optimization (MOCO) problems remains relatively unexplored. This paper presents a novel graph neural network (GNN) based DRL to configure multi-objective evolutionary algorithms. We model the dynamic algorithm configuration as a Markov decision process, representing the convergence of solutions in the objective space by a graph, with their embeddings learned by a GNN to enhance the state representation. Experiments on diverse MOCO challenges indicate that our method outperforms traditional and DRL-based algorithm configuration methods in terms of efficacy and adaptability. It also exhibits advantageous generalizability across objective types and problem sizes, and applicability to different evolutionary computation methods.
title Graph-Supported Dynamic Algorithm Configuration for Multi-Objective Combinatorial Optimization
topic Neural and Evolutionary Computing
Machine Learning
url https://arxiv.org/abs/2505.16471