Convergence Rate Maximization for Split Learning-based Control of EMG Prosthetic Devices

Employing SL is also suitable in scenarios featuring clients Centralized machine learning involves transmitting a vast with limited computational resources. The wearable devices amount of raw data, which may cause both increased latency utilized in both EMG-based prosthetic control and Human Activity and potential network congestion. Distributed learning Recognition (HAR) belong to this category. Specifically, techniques, such as Federated Learning and Split Learning, a significant issue caused by the low processing power of EMG actively address these issues by conducting distributed training controlled prostheses is their inability to incorporate large without transmission of raw data between the clients and the models. This property makes the application of deep learning server [1]. Therefore, these distributed learning approaches and FL impractical in such systems. Moreover, implementing are adequate for scenarios involving resource-constrained and small models does not satisfy the strict requirements regarding time-varying systems.