[coursera/dl&nn/week4]Deep Neural Network(summary&question)

Deep learning is an experiment base on hyperparameters.

I strongly encourage you to find a paper to write down forward and backward propagate.

You need to review how to compute the derivative of metrics.

Week4 Deep Neural Network

4.1 Deep L-layer neural network

n[0]=nx=3 ,  n[1]=5 ,  n[2]=5， n[3]=3， n[4]= n[L]=1

4.2 Forward Propagate in a Deep Network

We need a for-loop to compute these hidden layers.

compute lth layer neuron

vectorizing 

4.3 Getting your matrix dimensions right

This is important for us to modify our bug.

for  lth layer neuron in m training examples:

n[l]: num of neurons in lth layer

n[l-1]: num of neurons in (l-1)th layer

Z.shape()=(n[l],m)

W.shape()=(n[l],n[l-1])

X.shape()=(n[l-1],m)

b.shape()=(n[l],m)

4.4 Why deep representations

small: small amount of hidden nurons 

deep: large num of hidden layers

4.5 Building blocks of deep neural networks

The flow chat shows how deep neural networks propagate.

4.6 Forward and Backward Propagation

Backward:

vectorizing:

4.7 Hyperparameters

learning rate

#iterations

#hidden layer L

#hidden units

4.8 points: what does this have to do with the brain

mapping: x to y

question:

wrong answer:

1.the "cache" records values from the forward propagation units and sends it to the backward propagation units because it is needed to compute the chain rule derivatives.