Title: A Machine Learning Theory and Approach to Functional Data Analysis
Speaker: Prof. Xiao-Jun Zeng, University of Manchester
Time: 9:00 am, October 27, 2024
Venue: Room 109，Building No.12, Wushan Campus
Introduction to the speaker:
Dr Xiao-Jun Zeng is a Professor of Machine Learning in the Department of Computer Science at the University of Manchester, where he joined in 2002. Before joining to the University of Manchester, he was with Knowledge Support Systems (KSS) Ltd, UK between 1996-2002 as a scientific developer, senior scientific developer, and Head of Research, researching and developing intelligent pricing decision support systems for retail, petroleum, telecommunication, and finance industries.
Prof. Zeng’s main research areas include computational intelligence, machine learning, decision support systems, and their applications in retail, finance, and energy industries. His research has been funded UK EPSRC, Innovate UK, EU 6th and 7th Framework Programmes, and EU H2020 Programme, and has published more than 200 international journal and conference papers in these areas with more than 16000+ citations based on Google Scholar.
Prof. Zeng is an Associate Editor of the IEEE Transactions on Fuzzy Systems as well as several other journals. He received the European Information Society Technologies Award in 1999 and the Microsoft European Retail Application Developer Awards in 2001 and 2003 with KSS Ltd. His research in intelligent pricing decision support systems was recognised and selected by UK Computing Research Committee, Council of Professors and Heads of Computing, and British Computer Society Academy as one of 20 impact cases to highlight the impact made by UK academic Computer Science Research within the UK and worldwide over the period 2008 – 2013.
Abstract：
The world around us is continuous such as time, space, weather, and objects. Currently these continuous objects are represented by vectors and matrices as the digitalization technology we use. Although vectors and matrices as data representation are enough in many applications, in many other applications when objects being dealt with are continuous, vectors and matrices as data representation suffer some serious weaknesses such as information loss, overcomplicatingrepresentation (as continuous nature is not utilized), lower resolution prediction, and missing insights. To address such an important missing, functional data analysis (FDA), which deals with continuous data such as functions, curves, shapes etc., has been proposed and developed since 90s, mainly by the statistical research community but it is largely not being aware of or ignored by the machine learning/AI communities. To address such a gap, we have developed a machine learning theory and approach to FDA during the recent years and the talk will report our progresses in this topic.
