Dynamic Secure Modality-Aware Split Learning: A New Paradigm

Abstract

Split Learning (SL) has emerged as a promising approach for distributed machine learning, particularly in scenarios involving resource-constrained devices and privacy-sensitive data. This paper proposes a novel approach called Dynamic Secure Modality-Aware Split Learning (DSMASL), which adapts dynamically to device modalities while ensuring robust privacy measures. By integrating encryption mechanisms and dynamic adaptation techniques, DSMASL addresses the computational, communication, and privacy challenges identified in existing SL frameworks. Experimental results demonstrate the potential of DSMASL to significantly enhance the efficiency and security of distributed learning systems.