Udacity Google Deep Learning learning Notes

1. Why add pooling (pooling) to the convolutional network

If you only use convolutional operations to reduce the size of the feature map, you will lose a lot of information. So think of a way to reduce the volume of stride, leaving most of the information, through pooling to reduce the size of feature map.

Advantages of pooling:

1. Pooled operation does not increase parameters

2. Experimental results show that the model with pooling is more accurate

Disadvantages of pooling:

1. Because the stride of the convolution is reduced, the computational amount is increased

2. At the same time, the pooling layer allows us to add two extra parameters (hyper parameters): Pooling size and pooling stride

2. Why 1x1 convolution

The convolution of the input image is equivalent to a linear classifier, but if you add a 1x1 convolution between the input picture and the convolution layer, the two layers are equivalent to combining a small neural network (nonlinear).

Udacity Google Deep Learning learning Notes