5月17日 Hui Ji 教授学术报告

发布时间:2014-05-09

报告题目：Data-driven frames and tight frames for sparse modeling: theory, analysis and applications

报告人：Hui Ji, Associate Professor
（Department of Mathematics & affiliated with CWAIP，National University of Singapore）

时 间： 5月17日（周六） 上午10：00-11：00

地 点：旧数学楼212室

报告摘要：
Sparse modeling has been one main driving force of the recent development in image recovery and patter recognition. Most often used system for sparse presentation of image data are redundant systems, particularly frames and tight frames. In the first part of this talk, I will present a variational approch to construct wavelet-type tight frames that is adaptive to the input image for optimal sparse approximation. With comparable performance in image recovery, the proposed method is much faster than some representative sparsity-based dictionary learning method such as the K-SVD method.
In the second part, I will present a proximal linearized method for solving the non-convex L0 norm based optimization problems resulting from the first parts. The proposed solver differs from the existing ones by its theoretically justified global convergence, i.e., the sequence generated by the proposed method converges to a critical point of the corresponding non-convex problems. The practical benefit is also demonstrated in its applications in pattern recognition.

报告人简介：
http://www.math.nus.edu.sg/~matjh/

欢迎有兴趣的师生参加！