The Internet of Things (IoT) consists of a massive number of smart devices capable of data collection, storage, processing, and communication. The adoption of the IoT has brought about tremendous innovation opportunities in industries, homes, the environment, and businesses. However, the inherent vulnerabilities of the IoT have sparked concerns for wide adoption and applications. Unlike traditional information technology (I.T.) systems, the IoT environment is challenging to secure due to resource constraints, heterogeneity, and distributed nature of the smart devices. This makes it impossible to apply host-based prevention mechanisms such as anti-malware and anti-virus. These challenges and the nature of IoT applications call for a monitoring system such as anomaly detection both at device and network levels beyond the organisational boundary. This suggests an anomaly detection system is strongly positioned to secure IoT devices better than any other security mechanism. In this paper, we aim to provide an in-depth review of existing works in developing anomaly detection solutions using machine learning for protecting an IoT system. We also indicate that blockchain-based anomaly detection systems can collaboratively learn effective machine learning models to detect anomalies.

Split learning (SL) aims to protect user data privacy by distributing deep models between the client-server and keeping private data locally. In SL training with multiple clients, the local model weights are shared among the clients for the local model updates. This paper first reveals data privacy leakage exacerbated by local weight sharing among the clients in SL through model inversion attacks. Then, to reduce the data privacy leakage issue, we propose and analyze privacy-enhanced SL (P-SL) (or SL without local weight sharing). We further propose a parallelized P-SL to expedite the training process by duplicating multiple server-side model instances without compromising accuracy. Finally, we explore P-SL with late participating clients and devise a server-side cache-based training method to address the forgetting phenomenon in SL when late clients join. The experiment results demonstrate that P-SL helps reduce up to $50\%$ of client-side data leakage, which essentially achieves a better privacy-accuracy trade-off than the current trend by using differential privacy mechanisms. Moreover, P-SL and its cache-based version achieve comparable accuracy to baseline SL under various data distributions, while incurring lower costs for computation and communication. Additionally, caching-based training in P-SL mitigates the negative effect of forgetting, stabilizes learning, and enables practical and low-complexity training in a dynamic environment with late-arriving clients.