lecture 6 Neural Networks Part 3 Intro to ConvNets hd
outline:
SGD performes not so good in practice.
Momentum update ,物理解释很不错
Nesterov Momentum update
and so on…
二次优化的方法,Hessian矩阵要求逆,尤其是高维的时候,所以impratical
改进的
ensemble
dropout(regularization)
see the slides for detail..