# Pooling vs Subsampling: Multiple Definitions?

I have seen people using pooling and subsampling synonymously. I have also seen people use them as different processes. I am not sure though if I have correctly inferred what they mean, when they use the terms with distinct meanings. I think these people mean that the pooling part is selecting a submatrix from an input matrix and the subsampling part is selecting yet another submatrix that satisfies some condition from the first submatrix.

So say we have a $$100 \times 100$$ image. We do $$10 \times 10$$ non-overlapping pooling with $$5 \times 5$$ max subsampling. That would mean we slide the $$10 \times 10$$ "pool" across the image in strides of 10 and at every step we select the $$5 \times 5$$ submatrix inside the pool that has the maximum sum. That $$5 \times 5$$ matrix is what comes out of the $$10 \times 10$$ pool at the current potion. So in the end we have a $$50 \times 50$$ image.

Can you confirm that this usage of the terms pooling and subsampling exists?

I inferred this definition, as I cannot make sense of how some people use the two terms otherwise. For example in this video, or rather the people from the paper he is talking about (which I can't find because he only has the author name on his slide and no year).

• I think this is asking the same question: stats.stackexchange.com/questions/354944/… Jul 29, 2019 at 2:03
• @Recessive I had read that one. It does not cover the case where people use the terms not as synonyms. It does not explain what a "$10 \times 10$ pool $5 \times 5$ subsampling" is supposed to be. Jul 29, 2019 at 12:12
• From what I know, a maxpooling layer is a method for subsampling. So in that case, the terms are somewhat interchangable, but other types of pooling layers may also be a form of subsampling. Jul 30, 2019 at 4:09
• @Recessive The person from the video answered me now, and he says in this case by subsampling the authors meant the stride of the pooling. Jul 30, 2019 at 12:41