Critical Node Detection with Reinforcement Learning

Abstract

In this paper, we investigated how quickly graph connectivity can be weakened by removing nodes based on traditional importance metrics such as Closeness Centrality, Betweenness Centrality, and PageRank, compared to node removal based on Deep Reinforcement Learning (DRL) which generates the sequence of nodes in order of importance. By comparing the effectiveness of conventional importance metrics with those derived from DRL, the study examines the potential superior performance of Deep Reinforcement Learning in critical node detection.