Lei Chen has BS degree in computer science and engineering from Tianjin University, Tianjin, China, MA degree from Asian Institute of Technology, Bangkok, Thailand, and PhD in computer science from the University of Waterloo, Canada. He is a chair professor in the Department of Computer Science and Engineering, Hong Kong University of Science and Technology (HKUST). Currently, Prof. Chen serves as the director of Big Data Institute at HKUST and director of HKUST MOE/MSRA Information Technology Key Laboratory. Prof. Chen’s research interests include human-powered machine learning, crowdsourcing, Blockchain, graph data analysis, probabilistic and uncertain databases and time series and multimedia databases. Prof. Chen got the SIGMOD Test-of-Time Award in 2015. The system developed by Prof. Chen’s team won the excellent demonstration award in VLDB 2014. Prof. Chen has served as VLDB 2019 PC Co-chair. Currently, Prof. Chen serves as Editor-in-Chief of VLDB Journal, associate editor-in-chief of IEEE Transaction on Data and Knowledge Engineering. He is an IEEE Fellow, ACM Distinguished Member and an executive member of the VLDB endowment.

As one of the most basic meteorological and hydrological elements, rainfall plays an important role in disaster warning (i.e., floods and landslides). Providing accurate rainfall estimates in time is an important topic in many research fields. However, in practical applications, rainfall prediction (or estimation) is still an intractable problem due to the extremely complex spatiotemporal variability. Numerical models were devised to predict future precipitation and map rain gauges into high granularity, while the performance is still far from satisfactory. In recent years, deep learning has aroused more attention from many research fields such as image recognition, natural language processing as well as precipitation forecasting. In this talk, I will share our research experience on providing accurate rainfall estimates based on deep learning techniques, including rainfall forecasting and rainfall spatial estimation. For rainfall forecasting, we aim at the accurate prediction of radar images in HKO, which has shown a strong correlation with rainfall and has been used as an indicator of rainfall. Although rain gauges provide accurate measurements at certain locations, due to the data sparsity, it is quite challenging to conduct rainfall spatial estimation. We propose a Graph Neural Networks (GNN) approach to dynamically learn the spatial relationship of the rainfall.