Block-FeST: A Blockchain-Based Federated Anomaly Detection framework with computation offloading using Transformers
Zahra Batool, Kaiwen Zhang, Zhongwen Zhu, Sarang Aravamuthan, Ulrich Aïvodji
Abstract
Internet of Things (IoT) devices generate a massive amount of data on a regular basis. This data has the potential to revolutionize every sector by developing intelligent systems. Unfortunately, the data is inaccessible due to privacy concerns. A decentralized machine learning approach, i.e., Federated Learning (FL), enables clients to train models locally in an iterative and collaborative manner. However, some clients may not be able to fully participate in training due to limited computational resources. To address the aforementioned issues, we propose Block-FeST, a blockchain-based Federated-Split Learning framework for anomaly detection using Transformers, that incorporates the strength of Federated and Split Learning (FSL). FL is employed to mitigate data privacy issues, while Split Learning (SL) is used to offload some computational overhead from constrained clients to a central server. Block-FeST is also capable of training a transformer model, which we demonstrate in the context of a temporal anomaly detection (AD) problem. Moreover, the use of a blockchain will generate an audit trail that can be used to address any challenges from the customers to take corrective actions. We have implemented our proposed solution and compared it against known centralized and decentralized baselines. Block-FeST achieves an accuracy of 86%, which is competitive with the other solutions, while providing additional benefits in terms of decentralization and client-side offloading.