Data distribution aware clustering for parallel split learning in healthcare applications

Md. Tanvir Arafat, Md. Abdur Razzaque, Abdulhameed Alelaiwi, Md. Zia Uddin, Mohammad Mehedi Hassan

Abstract: Highlights•Introduces a clustering-based approach to improve split learning with non-independently distributed data.•Data distribution-aware clustering-based split learning (DCSL) optimizes client device clusters for faster convergence and reduced training latency.•Proposes a binary integer nonlinear programming formulation for clustering in split learning to handle data heterogeneity.•Develops a proximal policy optimization-based deep reinforcement learning method to solve the clustering problem.•DCSL outperforms existing methods in training accuracy and latency through simulations.

External IDs:dblp:journals/fgcs/ArafatRAUH26