随笔分类 -  Deep Learning

摘要：## [VGG](http://www.robots.ox.ac.uk/~vgg/publications/)- [Andrea Vedaldi](http://www.robots.ox.ac.uk/~vedaldi/)## [Berkeley](http://bvlc.eecs.berkeley... 阅读全文

摘要：- vanishing gradient problem- multi-dimensional lstm 阅读全文

摘要：Ref: 阅读全文

摘要：- `cross entropy` loss is not quite the same as optimizing classification accuracy. Althougth the two are correlated.- It's not necessarily true tha... 阅读全文

摘要：Andrew Ng 组的Tutorial做的很浅显易懂啊，今天明白了Autoencoder。其实，autoencoder做的就是降维，我觉得最让我眼睛一亮的地方是，用KL divergence(\ref{kl})做约束实现sparsity，相当于把$\rho$跟$\hat{\rho}$都看成是一种分布，统计的思想体现的很好，跟L1 norm有异曲同工之妙，而且，我感觉解析性应该比L1 norm更好！\begin{equation}\label{kl}\mathbf{ KL}(\rho || \hat\rho_j) = \rho \log \frac{\rho}{\hat\rho_j} + (1 阅读全文