At this point in a video on LeNet1, Yann LeCunn seems to make a distinction between pooling and subsampling, with a separate gesture for each:

[...] The second version had a separate convolution and pooling layer and subsampling

I believe I understand what max pooling and average pooling do. What is subsampling as a distinct operation?


I don't think there is any difference. Pooling operation does sub-sampling of the image. You can find that people refer to subsample as an operation performed by pooling layer

In fact, in the paper they describe sub-sampling as a pooling layer

You can check Yann LeCun's paper Gradient-Based Learning Applied to Document Recognition


There are different types of pooling, including MaxPooling and AveragePooling. MaxPooling captures the maximum pixel value in a grid (say z x z) from the entire image and then processes that in output image. Average Pooling likewise calculates the average and processes that in output image.

On the other hand, Subsampling chooses a pixel in the grid and replaces surrounding pixels of said grid by the same pixel value in the output image.

The output image from these two might look the same (and might have no major changes in accuracy in the Neural Network), however, they are not exactly same.

  • 1
    $\begingroup$ So then max pooling is subsampling with the maximum value selected. $\endgroup$ – Tom Hale Aug 3 '19 at 15:30
  • $\begingroup$ Exactly the point $\endgroup$ – Hmrishav Bandyopadhyay Aug 5 '19 at 13:28
  • $\begingroup$ In addition, sub-sampling produces trainable parameters whiles Max and Average pooling do not. Please refer to stats.stackexchange.com/a/265319/197343 $\endgroup$ – rocksyne Mar 5 at 10:33

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.