Pooling
Deep in the network after multiple Convolution layers the number of features quickly grows. To reduce the amount of features we can summarize them with a Pooling Operation.
It works like this:
Notice that the stride is usually bigger than with Convolution.
The example above shows one possible pooling function. But there are other commonly used functions:
- Maximum
- Average
- - Norm → all values squared and summed up.
Combining Convolution and Pooling can lead to a network looking like this:
