卷积神经网络吴恩达coursera

Convolutional NN

Foundations of CNN

matrixs convolution

Edge detection

Vertical / horizontial

conv-forward(tf.nn.cov2d)

$$matrix(6\times 6)\ast filter(3\times 3)=matrix(4\times 4)$$​

$$matrix(n\times n)\ast filter(f\times f)=matrix((n-f+1)\times (n-f+1))$$

Padding

problems:

• shrinking output
• throw away infomation from edge

$$matrix((n+2p)\times (n+2p))\ast filter(f\times f)=matrix((n+2p-f+1)\times (n+2p-f+1)$$

valid convolution / same convolution

“same” : $$p=\frac{f-1}{2}(oddf)$$

Stride

$$matrix((n+2p)\times (n+2p))\ast filter(f\times f)=matrix((\frac{(n+2p-f)}{s}+1)\times (\frac{(n+2p-f)}{s}+1)$$

Volume convolution

$$matrix(n\times n\times n_c)\ast filter(f\times f\times n_c)=matrix((n-f+1)\times (n-f+1)\times n_c)$$