Dr Shuo Wang, University of Birmingham

Millions of network devices are constantly generating a wealth of data. Distributed machine learning (ML) is necessary to split up learning tasks and enable parallel computation. However, data in such real-time systems is susceptible to changes over time, causing out-of-date models. Shuo Wang aims to bring adaptivity and improve robustness of distributed ML systems when network data is non-stationary. A specific and important type of data non-stationarity, correlation concept drift in multiple data stream scenarios, is studied in depth.