[ghayes]

Geoffrey Hinton had an excellent series on neural networks from 2011 for Coursera available here https://m.youtube.com/playlist?list=PLoRl3Ht4JOcdU872GhiYWf6... detailing the fundamentals of machine learning. The series was later wholesale replaced by another led by Andrew Ng of Google. I really adored Geoffrey’s lectures and recommend it to anyone looking to get into the space. It ends with him hinting at the idea of attention networks, but sadly I can’t find any later lectures from him on the topic.

[hintymad]

I took that course. His language was dense and was quite complex. It was like listening to someone reading out the Goodfellow's Deep Learning book, except that the language was even more dense. I guess it showed Hinton's amazing mental capacity

[rajnathani]

I totally agree, as I took that course during University of Toronto undergrad (either 2013 or 2014) when [Sir?] Geoffrey Hinton decided to flip (this is an official term) the course to have it that we watched the Coursera video course of what is linked above, and ask questions in class. It was extremely hard to understand the course material, and I dropped out a few weeks into it after the first unit test. It is probably one of the most awful introductions to neural networks that exists, and that's why the "go-to" ML courses are ones by Andrew Ng, Jeremy Howard's fast.ai, and others. But to be fair, the class was very math-heavy in terms of the actual underlying implementation of neural networks, and I seemed to be an exception in that class of about 50 or so students (many seemed from the master's program) who simply could grasp much of the math behind it (I'm sure anyone who understand the course material could implement neural nets at a CUDA level).