Policy-Aware Clouds: Federated Reinforcement Learning for Cross-Cloud Compliance

Abstract

In this paper, we study the use of Federated Reinforcement Learning (FRL) to resolve compliance issues among various multi cloud environments. However, due to growing concerns of data privacy and regulatory adherence, traditional centralized learning approaches are less suitable in distributed systems as well as in situations where data is not centralized. In this line, we propose a compliance aware FRL that allows learning optimal policies for cloud native agents without sharing sensitive data. The model provides high performance by enforcing regional compliance while maintaining policy convergence on the same level of simulated GDPR, HIPAA, and PCI-DSS governed clouds. Results show that FRL indeed can be a scalable, privacy preserving and regulation compliant solution for the modern enterprise cloud infrastructures.