top of page
Search
HAIV

ConvexLearning

凸学习算法


凸学习(Convex Learning)

凸光滑有界(convex smooth bounded)是针对一个假设H可学习性的一个定义,有关但有别于凸函数的定义(具体到凸优化算法,我们后面会涉及凸损失函数)。一个凸光滑有界的学习问题一定是可学习的


随机梯度下降(SGD)

对于凸光滑有界的学习问题,我们有

算法如下:

梯度下降(Gradient Descent)

梯度下降的推导来源于泰勒展开。算法如下:

不可导函数情形

如果函数不可到,可以计算sub-gradient

相应地,更新一步修改为

其中


40 views0 comments

Recent Posts

See All

Random Feature Fourier and Double descend

Prof. Zhenyu Liao from School of Electronic Information visited HAIV Lab and gave a talk on the double descend effect of training loss...

Random Features

On April 2, 2021, Dr. Fanghui Liu visited HUST and gave a talk at HAIV Lab on random feature Fourier for kernel approximation. He is...

MetaLearn

Comments


bottom of page