Statistical Machine Learning via Distribution Comparison

Speaker: Wenkai Xu, Foundations of Machine Learning System, Tuebingen AI Center, University of Tuebingen, Germany

Abstract: 

Machine learning methods have enjoyed evolutionary success in recent years. A natural question to ponder is how the machine actually "learns"? 

In this talk, we address this question by discussing the statistical nature of machine learning. We first introduce the concepts and tools for distribution comparisons. We show examples of commonly used measures of distributions in machine learning to emphasise the crucial role of loss function in learning procedures. We then illustrate various applications based on distribution comparison, such as hypothesis testing in scientific discovery, variational inference for image processing, as well as deep generative modelling. We conclude the talk by addressing the social impacts derived from statistical machine learning and potential problems to be guard against. We give examples of privacy-preserving learning problems and fairness-aware learning objectives.

Biography: 

Wenkai is a Postdoctoral Fellow at the Foundations of Machine Learning System (FMLS), Tuebingen AI Center and the University of Tuebingen. Prior to that, Wenkai worked as a postdoc research associate in Statistics at the Department of Statistics, University of Oxford and an elected research associate at Keble College, University of Oxford. He completed his PhD in computational neuroscience and machine learning at Gatsby Computational Neuroscience Unit, University College London.
Wenkai's research mainly focuses on statistical hypothesis testing, information processing, and Stein's method for complex data. In particular, he aims to continue to investigate useful properties of statistical measures between distributions for developing practical tools in machine learning systems. Moreover, he is also interested in the development and adjustment of machine learning systems to address issues pertaining to privacy, fairness and ethics.

Host: Prof. Ernst-Jan Camiel Wit

Click here to join online