Convolutional Neural Network Notes

Notes:

1. Dimensions:

• Input layer: $(W \times H \times D)$
• $K$ filters $(F \times F \times D)$
• Stride $S$ and padding $P$

$\rightarrow$ Width & height of the output layer:

$\rightarrow$ Shape of output layer: $(W_{out} \times H_{out} \times K)$

$\rightarrow$ Number of parameters: $(F \times F \times D + 1) \times K$

2. Pooling

Benefits of pooling:

• Reduce the size of the input
• Prevent overfitting

Max pooling:

• Allow the neural network to focus on only the most important elements