Privacy-Preserving Anomaly Detection for IoT: Leveraging Federated and Split Learning
Arjdal Rguibi, Asimi Younes, Ahmed Asimi, Lahcen Oumouss · 2024
The abundance of Internet of Things (IoT) devices calls for resilient security measures capable of detecting abnormal activities that signal possible cyber threats. Nevertheless, conventional centralized methodologies frequently encounter difficulties associated with data confidentiality and limitations in resources on singular devices. This paper delves into the utilization of federated learning and split learning for detecting anomalies in IoT security. Federated learning enables collaborative model training on distributed devices without compromising data privacy. Split learning further distributes the training process by keeping a portion of the model on each device to address resource limitations. We discuss the design considerations for federated and split learning models tailored to anomaly detection in IoT data. Our research focuses on preserving privacy, handling resource constraints, and adapting to evolving normal data patterns. This work contributes valuable insights into leveraging federated and split learning for secure and scalable anomaly detection in the context of IoT security.