Efficient ADMM and Splitting Methods for Continuous Min-cut and Max-flow Problems

报告人及简介：

孙鸿鹏，中国人民大学副教授。2012年博士毕业于中科院数学与系统科学研究院，2012年-2014年底于奥地利Graz大学做博后，2015年入职中国人民大学，2016年曾获德国洪堡基金会资助。研究方向为反问题和图像处理，与合作者的工作曾发表于 SIAM Journal on Numerical Analysis，JMPA, Inverse Problems等著名国际期刊。

报告时间：2020年10月13日，下午3:00-4:00

ZOOM会议 ID：982 9447 9333，会议密码：888888

会议链接：https://zoom.com.cn/j/98294479333?pwd=U1J5ZGVtcHo2OXlaZE5zbjlid3U4Zz09

摘要：The Potts model has many applications. It is equivalent to some min-cut and  max-flow models. Primal-dual algorithms have been used to solve these  problems. Due to the special structure of the models, convergence proof  is still  a difficult problem. In this work, we developed two novel,  preconditioned, and over-relaxed alternating direction methods of  multipliers (ADMM) with convergence guarantee for these models. Using  the proposed preconditioners or block preconditioners, we get  accelerations with the over-relaxation variants of preconditioned ADMM.  The preconditioned and over-relaxed Douglas-Rachford splitting methods  are also considered for the Potts model. Our framework can handle both  the two-labeling or multi-labeling problems with appropriate block  preconditioners based on Eckstein-Bertsekas and Fortin-Glowinski  splitting techniques. This is joint work with Prof. Yuan, Jing and Prof.  Tai, Xuecheng.