Home // ICAS 2025, The Twenty-First International Conference on Autonomic and Autonomous Systems // View article

Applying Q-Learning Agents to Distributed Graph Problems

Authors:
Jeffrey McCrea
Munehiro Fukuda

Keywords: agent-based modeling; cluster computing; graph algorithms

Abstract:
Breadth-first search is used as a brute-force approach to parallelizing graph computations over a distributed graph structure, such as the shortest path, closeness centrality, and betweenness centrality search. As a smart alternative, we integrate Q-learning capabilities into agents, dispatch them over a distributed graph, have them populate the Q-table, and accelerate their graph computations. We developed the Q-learning agents with the Multi-Agent Spatial Simulation (MASS) library and measured their parallel performance when running over a distributed graph with 16K or more vertices. This paper identifies the graph scalability, static/dynamic graph structures, application types, and Q-learning hyperparameters that take advantage of Q-learning agents for parallel graph computing.

Pages: 29 to 34

Copyright: Copyright (c) IARIA, 2025

Publication date: March 9, 2025

Published in: conference

ISSN: 2308-3913

ISBN: 978-1-68558-241-8

Location: Lisbon, Portugal

Dates: from March 9, 2025 to March 13, 2025