The main idea

1. Reduce the number of input nodes.

2. Tolerate small shifts in where the pixels are in the image.

3. Take advantage of the correlations that we observe in complex images.

Algorithm

1. Input

2. Filter

3. Feature map

4. Run the feature map through a Relu activation function, then all the negative values are set to 0, and the positive values are the same as before.

5. Apply a new filter to the feature map:

  1) Max pooling: simply select the maximum value.

  2) The filter moves in such a way that it does not overlap itself.

  3) Average/ mean pooling:

Terminology

Filter: a square that is commonly 3 pixels by 3 pixels

Feature map:  value = input * filter + bias, slide the filter 1 pixel. Help take advantage of the correlations.