Influence Maximization in Dynamic Networks Using Deep Reinforcement Learning

Abstract

Influence maximization (IM) seeks to maximize the spread of influence across networks, a problem with critical applications in social media,
viral marketing, and epidemic control. While extensively studied in static networks, the dynamic nature of real-world networks presents unique challenges, including varying network structures and computational efficiency. In this seminar, we explore a recently proposed approach to IM in dynamic networks by Dizaji et al. (2024), which uses Deep Q-Learning (DQL) to address these challenges. This method integrates incremental and transfer learning techniques to adapt
to evolving network features while retaining learned knowledge.
Experimental results demonstrate the efficacy and scalability of this method, outperforming traditional techniques on both synthetic and real-world datasets. 

The talk will highlight the methodology, challenges addressed, and potential applications of this framework.