设为首页/加为收藏 /北邮主页/ 党委宣传部/ 联系我们
通知公告
您当前所在的位置: 网站首页 -> 通知公告 -> 正文

理学院青年学术论坛第197期——High-order multilevel optimization strategies and their application to the training of ANNs

发布日期:2019-03-28  来源:理学院   点击量:

主讲人:Geovani Nunes Grapiglia博士

时间:2019年4月9日10:00-11:00

地点:主楼1214

邀请人:孙聪

报告题目:Improved Optimization Methods for Image Registration Problems

报告摘要:

In many applications, multiple images of the same subject are obtained at different moments in time and under different conditions. The goal is to identify changes in the subject over time. For a fair comparison, the images need to be aligned in order to overlap the common features. Image Registration is the problem to find the transformation of images leading to the best overlapping. The mathematical modelling of this type of problem often results in large-scale unconstrained optimization problems to solve. The usual approach is to apply Gauss-Newton Method with Armijo line-search embedded in a multi-level coarse-to-fine scheme, where the solution of the coarse level is used to generate the initial point for the fine level. In this talk, two alternative approaches will be discussed, including the use of L-BFGS and subspace techniques.

主讲人介绍:

Dr. Geovani Nunes Grapiglia got his phD degree in Mathematics in 2014 from Federal University of Parana, Brazil. He is now an assistant professor in Federal University of Parana. During the year 2013, he visited Chinese Academy of Sciences supported by the "Sandwich Doctorate Scholarship". He got the SIAM student travel grant in 2014. His research interests are nonlinear optimization theory and algorithms, particularly trust region methods, derivative-free methods, Newton's method and error bound analysis. He has published several papers on Math Programming, SIAM Optimization, and so on.

热点新闻

电子信箱:news@bupt.edu.cn   新闻热线:010-62282031      通讯地址:北京市海淀区西土城路10号    邮编:100876

Copyright ? 2018  All Rights Reserved  北京邮电大学党委宣传部(新闻中心) 版权所有

京ICP备 05064445号    京公网安备110402430070

香港内部正版资枓大全