Abstract:

Statistical dependence is the main driving force for current intelligent learning systems. With the learned dependence, one can estimate beliefs or probabilities when the data generating process is fixed. However, we humans are able to adapt ourselves to new situations and make accurate predictions and decisions, because we have a good causal understanding of the world. In this talk, I will present how we can equip machines with causal learning capabilities, so that they can predict the effects of interventions and transfer knowledge to new environments. Specifically, I will introduce how we can infer causal relations from observational data, such as time series. In addition, I will discuss how to use causal models to tackle transfer learning problems in which the data distributions change across domains.

Bio:

Mingming Gong is a Lecturer (Assistant Professor) in data science with the School of Mathematics and Statistics, University of Melbourne. His research interests include causal inference, machine learning, and computer vision. He has authored and co-authored 30+ research papers including NeurIPS, ICML, UAI, AISTATS, CVPR, ICCV, ECCV, and AAAI. He has studied how the causal generative process of data benefit learning in non-standard settings, such as transfer learning and weakly-supervised learning. He also studies principles and methods to infer causal models from various kinds of observational data, including under-sampled time series, data with measurement error, nonstationary/heterogeneous data.