Split Learning-Enabled Framework for Secure and Light-weight Internet of Medical Things Systems

TL;DR

This work tackles IoMT malware detection on resource-limited devices by deploying a split learning framework that partitions a CNN between IoMT clients and an edge server to reduce on-device computation while preserving privacy. It introduces a game-theoretic joint optimization to balance energy consumption and communication latency, parameterized by $\alpha \in [0,1]$, and demonstrates superior performance over Federated Learning in accuracy, F1-score, convergence speed, and resource use. Empirical results show improvements of $+6.35\%$ in accuracy and $+5.03\%$ in F1-score, along with faster convergence and notably lower client-side computation and communication costs, validating SL as a scalable, secure paradigm for IoMT security. The framework also provides actionable insights into cut-layer placement, number of participating clients, and the trade-offs between energy and latency, making it adaptable to diverse IoMT hardware constraints and network conditions.

Abstract

The rapid growth of Internet of Medical Things (IoMT) devices has resulted in significant security risks, particularly the risk of malware attacks on resource-constrained devices. Conventional deep learning methods are impractical due to resource limitations, while Federated Learning (FL) suffers from high communication overhead and vulnerability to non-IID (heterogeneous) data. In this paper, we propose a split learning (SL) based framework for IoT malware detection through image-based classification. By dividing the neural network training between the clients and an edge server, the framework reduces computational burden on resource-constrained clients while ensuring data privacy. We formulate a joint optimization problem that balances computation cost and communication efficiency by using a game-theoretic approach for attaining better training performance. Experimental evaluations show that the proposed framework outperforms popular FL methods in terms of accuracy (+6.35%), F1-score (+5.03%), high convergence speed (+14.96%), and less resource consumption (33.83%). These results establish the potential of SL as a scalable and secure paradigm for next-generation IoT security.

Figures (5)

• Figure 1: Proposed Split Learning-based Framework for IoMT Malware Detection
• Figure 2: Comparing loss, accuracy, and F1-score of the proposed SL framework with the baseline FL models across epochs, average client processing times, and total client processing TFLOPs
• Figure 3: Comparing accuracy and F1-scores of the proposed SL framework for different split positions across epochs, average client processing times, and total client processing TFLOPs
• Figure 4: Comparing performance of proposed split learning framework based on the number of clients participating in the feed forward network
• Figure 5: Energy–latency trade-offs under different system constraints and weight settings $(w_1,w_2)$.