Data sketching in engineering sensor networks

When analysing data at scale, accurate solutions become computationally cumbersome demanding substantial memory and processing power. In situations where time is critical, e.g. for controlling a process, detecting an anomaly or predicting failure from incoming sensor network data, an approximate answer to the right problem often suffices. To this end, this project will explore how we can analyse such data without resorting to storing an expanding dataset using sketching algorithms based on randomised linear algebra suitable. The intension will be to develop tools for real-time diagnostics using streaming data from avionic or automotive sensors.