A Privacy-Preserving Multi-Objective Reinforcement Learning Framework for Scalable Big Data Knowledge Discovery and Distributed Processing

Abstract

This paper introduces a new paradigm that builds upon the Pareto Q-learning that supports multi-objective reinforcement learning to deal with privacy preservation and scalability of big data knowledge discovery and distributed processing. The suggested design incorporates deep function approximation methods that are effective at addressing big and continuous state action spaces typical in the big data analytic conditions of reality. Federated learning and DP are privacy-preserving systems which are incorporated to protect sensitive information during distributed computation. New exploration-exploitation strategies are presented to solve the privacy-conscious, large-scale data environment to boost learning speed and efficiency. The framework also uses adaptive set assessment and pruning algorithms to cope with the complexities in the computations required to maintain Pareto optimal sets of policies. Large-scale experiments on benchmark data sets show that the framework has high ability to balance between various objectives of privacy, scalability, and accuracy of knowledge extraction when compared to the state-of-the-art multi-objective reinforcement learning methodologies. The findings justify the method being proposed as scalable, secure and effective in terms of privacy aware big data mining using distributed multi objective reinforcement learning.