Nonsmooth Nonconvex Optimization for Training Deep Neural Networks

来源:数学学院  作者:王承竞     日期:2022/6/20 14:39:10   点击数:1538  

时间:2022-06-24 13:45

地点:腾讯会议会议号:885 416 848;密码:0624



This presentation includes 4 parts.     

1.  Reformulations of nonsmooth nonconvex optimization for training leaky ReLU neural network    

2.  Penalty methods and a smoothing proximal gradient algorithm    

3.  Computation of subgradients in stochastic gradient algorithms for nonsmooth nonconvex finite-sum optimization    

4.  Automatic choroid layer segmentation from OCT images using deep Learning