FedCST: Federated Learning on Heterogeneous Resource-constrained Devices Using Clustering and Split Training

Wang, Zhilu; Lin, Haiyang; Liu, Qi; Zhang, Yonghong; Liu, Xiaodong

Abstract

With the rapid development of 5G and Internet of Things (IoT) technologies, edge devices such as sensors, smartphones, and wearable devices have become increasingly prevalent. The massive amount of distributed data generated by these devices offers unprecedented opportunities for deep learning, especially in fields like computer vision and edge computing. However, this data often exists in isolated data silos, and due to its privacy-sensitive nature, it cannot be directly processed in a centralized manner. Additionally, the data heterogeneity and devices heterogeneity further exacer-bates the challenges in federated learning (FL). Moreover, with the implementation of data privacy protection regulations, user demand for privacy protection is growing significantly. Therefore, how to effectively utilize dispersed data for distributed collaborative learning while ensuring data privacy is a pressing issue that needs to be addressed in the field of FL. In response to these challenges, this work has conducted research on clustering and segmentation FL algorithms for heterogeneous resource-constrained devices. A lightweight homogeneous device clustering strategy has been designed, which incorporates a split learning (SL) mechanism to enhance the accuracy and training efficiency of FL models. This approach reduces the load on resource-constrained devices and improves privacy security. Consequently, this method can meet the dual requirements of privacy protection and computational efficiency in real-world scenarios.

Citation

Wang, Z., Lin, H., Liu, Q., Zhang, Y., & Liu, X. (2024, July). FedCST: Federated Learning on Heterogeneous Resource-constrained Devices Using Clustering and Split Training. Presented at The 24th International Conference on Software Quality, Reliability, and Security Companion (QRS-C), Cambridge, UK

• Presentation Conference Type: Conference Paper (published)
• Conference Name: The 24th International Conference on Software Quality, Reliability, and Security Companion (QRS-C)
• Start Date: Jul 1, 2024
• End Date: Jul 5, 2024
• Acceptance Date: May 9, 2024
• Online Publication Date: Oct 29, 2024
• Publication Date: 2024
• Deposit Date: Feb 10, 2025
• Publisher: Institute of Electrical and Electronics Engineers
• Peer Reviewed: Peer Reviewed
• Pages: 786-792
• Series ISSN: 2693-9371
• Book Title: 2024 IEEE 24th International Conference on Software Quality, Reliability, and Security Companion (QRS-C)
• ISBN: 9798350365665
• DOI: https://doi.org/10.1109/qrs-c63300.2024.00105
• Public URL: http://researchrepository.napier.ac.uk/Output/4113786