Effective Segmentation Learning Techniques for Dealing with Small Training Data
数学专题报告
报告题目(Title)：Effective Segmentation Learning Techniques for Dealing with Small Training Data
报告人(Speaker)：Ke Chen (University of Strathclyde, UK)
地点(Place)：后主楼1223
时间(Time)：2023 年 12 月 25 日（周一） 9:00--10:00
邀请人(Inviter)：段玉萍，刘君
报告摘要
Deep learning has quickly centered into all fields of research, though many real life problems are still too demanding and remain to be solved.
This talk discusses the topic of robust and accurate image segmentation. We start from the variational modelling approach and end up with the data modelling approach. Image segmentation for each broad class of images may be satisfactorily done quickly and reliably when the number of trained data is large enough. This is consistent with the theory available for image classification: accuracy is correlated with size of trained data.
For some application areas such as from medical MR images with inhomogeneous intensity or digital pathology with complicated patterns, the problems are it is either time-consuming to acquire enough trained data or impossible to get any at all. We present our preliminary work on developing and using variational models to assist the augmentation of training data and hence to increase the accuracy of learning methods or to facilitate the learning process that is otherwise not possible.
主讲人简介
Professor Chen is an applied and computational mathematician. His interests are in developing and analysing new and novel algorithms for a range of scientific and engineering applications. He has collaborated with many UK industries on mathematical modelling problems. More than 10 of his previous PhD students worked directly with industrial funding. His current interests are in developing imaging analysis techniques for high resolution image processing problems (mainly inverse problems) using a range of mathematical tools such as variational models, PDEs, iterative solvers and deep learning methods. He has paid particular attention to medical imaging problems (such as segmentation and co-registration), engaging directly with medical doctors. Currently he holds an honorary position at Clatterbridge Cancer Centre (NHS) as a clinical consultant. Prior to joining Strathclyde, he was the director of two multidisciplinary research centres (CMIT and LCMH) in the University of Liverpool.
