Optimization of trusted wireless sensing models based on deep reinforcement learning for ISAC systems

Abstract

This paper investigates using deep reinforcement learning (DRL) methods for optimizing trustworthy federated learning models, with a focus on integrated sensing and communication in practical wireless sensing scenarios. Challenges include computational disparities among edge sensing nodes, network transmission differences, and the non-independent and identically distributed (non-IID) nature of local training datasets. As the number of edge sensing nodes increases, the likelihood of encountering untrusted nodes also rises, further limiting the performance of traditional federated learning aggregation algorithms. To address these issues, the paper proposes a DRL-based strategy aimed at optimizing the node selection process in federated learning environments. This strategy intelligently selects nodes for global aggregation, improving overall model performance and efficiency by addressing computational and communication differences among nodes and the non-IID nature of data.