Calculation chart of classic two-layer-net,including forward & backward process.

• s: batch size,the number of samples in a batch.
• i: input size,the number of neurons in the input layer.
• h: hidden size,the number of neurons in the hidden layer.
• o: output size,the number of neurons in the output layer.

• For any param x in the forward process,dL/dx in the backward process has the same shape of it.
• During the backward process of the sigmoid layer,the mutiplication object of the martrix is elements of the martrix rather than the whole martrix.(not dot()).