Hmm, so I'm reading The Hundred Page Machine Learning Book [0] which gives an overview of the field, and one of the sections is on convolutional neural networks, and it was very crisp and clear (like the whole book, really; highly recommended). I understand since it's a brief summary it's probably eliding stuff, but it's so much shorter and more understandable than this blog post that it makes me wonder if it's really eliding a lot of crucial stuff, or if this post is just... not clear.

In the book, it walks you through the convolution of an input matrix with a smaller filter matrix. You start with lining up the smaller matrix in the top left corner of the input matrix, and you multiply each overlapping square together and add them all up. That's the value of the top left element of the result (the result is a matrix). Then you slide the filter to the right, repeat the calculation, and that's the value one to the right in the result matrix. You repeat, panning and scanning across the input matrix and in this fashion fill in the the whole result matrix.

I thought I had it, but this blog post just confused things. I do remember learning about convolutions back in undergrad and being confused then, too, so that sounds about right. When I read the ML handbook, I had a "wait, is that it?" thought and was confused why I was confused back in the day. But I guess that's just one particular way of doing it, or one use case or something.

[0] https://www.amazon.com/Hundred-Page-Machine-Learning-Book/dp...

I did bot have a good experience with Goodfellow’s Deep Learning Book. It’s much more written for the scholar than the practitioner. Furthermore, a couple relatively important methods (Transformer Layers, Causal convolutions, etc.) aren’t given proper treatment because the field was still moving fast at the time of publishing.

A much better investment would be the One Hundred Page Machine Learning Book. It’s written with practitioners in mind. Link here: https://www.amazon.com/Hundred-Page-Machine-Learning-Book/dp...: