Deep convolutional framelets: a general deep learning framework for inverse problemsMS65

Recently, deep learning approaches with various network architectures have achieved significant performance improvement over existing iterative reconstruction methods in various inverse problems. However, it is still unclear why these deep learning architectures work. Here we show that the long-searched-for missing link is the deep convolutional framelets expansion for representing a signal by convolving local and non-local bases using multi-level decomposition. Using numerical experiments with various inverse problems, we confirm the validity of our discovery.

This presentation is part of Minisymposium “MS65 - Machine learning techniques for image reconstruction (2 parts)
organized by: Markus Haltmeier (University Innsbruck) , Linh Nguyen (University of Idaho) .

Jong Chul Ye (Department of Bio and Brain Engineering, Korea Advanced Institute of Science and Technology)
Yoseob Han (KAIST)
Eunju Cha (KAIST)
deep learning, framelets, hankel matrix, image reconstruction, inverse problems, machine learning