Meta-Optimized Cooperative Autonomous Infrastructure Framework for Predictive, Energy-Aware, and Trustworthy AI Cloud Management

Abstract

Managing cloud infrastructure has become more challenging as the number of artificial intelligence (AI) workloads are increasing. In this regard, reactive, energy-inefficient, and uncooperative rule-based traditional orchestrations and single-agent reinforcement learning are ineffective. To address these issues, we propose in this paper a meta-optimized cooperative autonomous infrastructure (MOCAI) to enable predictive, self-managed, and trustworthy AI management. Our approach is based on prediction (using a temporal graph transformer), multi-agent meta-reinforcement learning, multi-agent negotiation (using games), trust-sized consensus, and explainable governance of decision-making. MOCAI is different from before, as it is a resource optimization system that predicts infrastructure behavior and cooperation of agents via cooperative intelligence. The system was tested in a cloud-edge system, which is 92.4 percent resource-efficient with 21 percent energy savings, 1.9 percent service level agreement (SLA) breach, and 95.6 percent secure. The performance of the prediction and cooperating negotiation strategies has been proven in the ablation studies. This research will take the autonomous infrastructure to the next level of scalable, secure, and sustainable artificial intelligence (AI) ecosystems, which can be used for the next generation of smart cloud ecosystems.