- used to reduce dimension - also has hyperparameters $f$ ![[CleanShot 2024-06-27 at [email protected]|400]] - there's max pooling & average pooling, where you either take the max of all elements, or mean of all elements - same output dimension as convolution layer: $\lfloor\frac{n + 2p - f}{s} + 1\rfloor\,X\,\lfloor\frac{n + 2p - f}{s} + 1\rfloor$ - but no parameters to learn unlike convolution layer! a fixed function - very rarely would you use [[Padding]] here