Architectural Enhancement of Environmental-Aware Federated Inference Orchestration for Sustainable AI/ML Management

Sustainability becomes one of the essential components of the future mobile network. In the meantime, computationally starving AI/ML applications are putting increased demands on energy consumption. In this paper, from the perspective of AI/ML orchestration, we propose a concept of environmentalaware federated inference, which utilizes the computing resources of network host nodes for the AI/ML model inference. Moreover, an enhancement on AI/ML orchestration is proposed that enables the federated inference to deploy AI/ML inference on network host nodes in a way that minimizes carbon emissions. Specifically, the enhancement includes new information elements required for the orchestration and detailed orchestration procedures in the mobile network. The architectural enhancement presented in this paper enables the AI/ML model deployment with minimal carbon footprint and provides reference for future sustainability related standardization activities. Index Terms—Mobile Network Architecture