BIPPO: Budget-Aware Independent PPO for Energy-Efficient Federated Learning Services

BIPPO represents a significant advancement in federated learning (FL), particularly within the context of large-scale IoT systems where resource constraints are prevalent. Traditional FL methods often overlook infrastructure efficiency, leading to challenges in client selection and overall performance. BIPPO, or Budget-aware Independent Proximal Policy Optimization, fills this gap by utilizing a multi-agent reinforcement learning approach that not only enhances accuracy in image classification tasks but also operates within a minimal budget. Evaluated on two distinct tasks, BIPPO demonstrated superior performance compared to non-reinforcement learning mechanisms and traditional methods like PPO and IPPO. This improvement is crucial as it ensures that FL can be effectively implemented in environments where resources are limited, thus promoting sustainability and efficiency in machine learning applications.