Abstract

Internet of Things (IoT) ubiquitous sensors and devices are generating massive data streams continuously. These streams need to be processed on-the-fly to extract knowledge for several applications like video surveillance, autonomous vehicles, smart city, web monitoring, etc. The existing approach for data stream processing is designed for centralised systems where all the data is sent to the data centres for storage and analytics. However, it is often not feasible to migrate all the data to the cloud for cost, performance and privacy concerns. In distributed systems like IoT networks, other agents like end devices, edge nodes, and cloudlets can cooperatively participate in the processing pipeline. This talk will focus on the design and deployment of deep learning algorithms on distributed nodes to tackle the challenges of data stream processing in distributed systems. We will explore how these algorithms can be optimised to meet system requirements in terms of scalability, low-latency and resource constraints. The potentials of deep autoencoders for data preprocessing on the edge using dimensionality reduction, anomaly detection and clustering techniques will be presented.