七分钟带你读懂深度学习

Forward-prop

Weight/Basis

SVM/logistic regration - simple pattern

supervised learning - labelled data 

text/speech processing - RNN (Recurrent Net)

image processing - CNN (Convolutional Net)

backpropagation

vanishing gradient

Bloltzmann machine: RBM ->feature extractor/autoencoder

Deep Belief Network (DBN)

supervised learning

Convolutional layer

Filter

BELU

Pooling

Fully connected layer (FC) - classify data samples

supervised learning method- need labelled data for training

Gating - LSTM/GRU

autoencoder - RBM

backpropagation - loss

use autoencoder to reduce the dimensionality of data

autoencoder is better than PCA

Recursive Neural Tensor Networks (RNTN)

Sentiment analysis

MRI, CT scan