Many problems of medical image analysis and machine learning are challenging due to the associated complex optimization formulations and constraints, extremely big image data being processed, poor imaging quality, missing data etc. On the other hand, it is highly desired to process and analyze the acquired imaging data, for example segmentation and registration etc., in an automated and efficient numerical way, which motivated vast active studies during the last 30 years, in a rather broad sense. This talk targets to present an overview of modern dual optimization theory, which delivers an advanced unified framework of mathematical analysis and develops high-performance numerical schemes with a wide spectrum of applications. We focus on the optimization problems arising from the most interesting topics: segmentation, registration and machine learning, and present both analysis and high-performance numerical solutions in a unified manner in terms of dual optimization.

Jing Yuan is working as the professor at the School of Mathematics and Statistics, Xidian University in Xi'an, China. Before that, he worked as the research scientist at Robarts Research Institute of Western University in Canada for over five years, and meanwhile the adjunct research professor at the Medical Biophysics Department of Schulich Medical School, Western University. He obtained his PhD with excellence from the Department of Computer Science and Mathematics in Heidelberg University, Germany. His research interests are in developing novel optimization theories and algorithmic implementations, advanced variational analysis and high-performance distributed parallel computing, especially with applications to most challenging practices of computer vision, medical image analysis and machine learning. He published over 80 papers in the top international journals and conferences, and is serving as the committee member or reviewer of many top conferences and journals.