Efficient Vanilla Split Learning for Privacy-Preserving Collaboration in Resource-Constrained Cyber-Physical Systems

In the realm of Cyber-Physical Systems (CPS), the integration of Federated Machine Learning (FML) algorithms has become indispensable for enhancing adaptability, data privacy, and security. However, FML falls short when deployed in resource-constrained CPS environments due to its inherent demands on client resources. To address this limitation, this paper introduces a novel Split Machine Learning (SML) architecture tailored specifically for resource-constrained CPS deployments. Unlike FML, SML strategically splits the model between devices and a central server, enabling collaborative learning while preserving data privacy. Adopting a distributed learning paradigm, SML facilitates real-time system adaptation based on local sensor data, mitigating communication overhead and ensuring privacy. Experimental evaluation demonstrates that the proposed SML-based architecture achieves an accuracy rate of 97.56% with a processing time reduction of approximately 41% compared to FML methods. These results highlight the potential of our approach to improve collaboration in resource-constrained environments while maintaining high levels of privacy and performance.