supervised learning
由一组参考的输入输出x、y， mapping得到相应的关系y=f（x）
structure data
data indatabase例如年龄、房价预测中的zip code、bedrooms、wealth等
unstructure data
audio、image、text，
standard NN
RNN
更适合one-dimension structured data 例如语言翻译machine translation、audio
CNN
photo tagging
week2
loss function
measure how good your prediction is（on a single training example ），L(y-hat，y)
coss function
on entire training set
learning rate