报告题目：A Convergent Learnable Optimization Algorithm for a Class of Nonsmooth Nonconvex Inverse Problems

主 讲 人：陈 韵 梅

单 位：佛罗里达大学

时 间：8月17日8:45

ZOOM ID：210 089 8623

密 码：123456

摘 要：

We propose a general learnable optimization framework for solving nonsmooth and nonconvex inverse problems, where the regularization function is the L_{2,1}-norm of a smooth but nonconvex feature mapping parametrized by a deep convolutional neural network. The proposed algorithm is a gradient decent type method that combines Nesterov’s smoothing technique and idea of residual learning, uses an iterate selection policy and adaptively reducing the smoothing factor to guarantee the convergence. Our method is versatile as one can employ various modern network structures into the model, and the resulting network inherits the guaranteed convergence of the algorithm. We also show that the proposed network is parameter-efficient, and its performance compares favorably to the state-of-the-art methods in a variety of image reconstruction problems in practice.

简 介：

陈韵梅，佛罗里达大学终身教授、杰出教授。致力于数学、图像处理和机器学习等交叉学科的研究，研究领域涉及医学图像分析中数学模型的建立与数值优化方法的发展，并对其中潜在的数学理论进行了深入的研究。曾获中国国家自然科学三等奖和教育部科技进步一等奖，获国际发明专利9项，主持国家级项目20余项，在Inventiones mathematicae, SIAM Journal on Imaging Science等杂志上发表学术论文200余篇。