Thank you. I overlooked this. My bad. I did not watch all the lectures because I do not quite get what category theory can do that can’t be done without it. Maybe I should be more patient.
Here’s the start of the type checker section:
I suppose it isn’t required to know category theory to find these results. But I do believe it will be a useful tool to have in general.
I just re-watched a few of their lectures almost in full, but I can’t quite follow the flow in the lectures. For example, in lecture 2, at 1:28:51, the speaker said a cylinder is a carrier object to deliver a message where input is a circle and a line and then the output is a line. I have no idea what he is saying. What is the meaning of a cylinder carrying a circle and a line and it becomes a line?
On the other hand, it seems to me what they are trying to do is, they can represent concepts in graphs, which is like sometimes a company would like to draw their business processes in some flow charts to list out all the intermediate steps. It’s nice to organize information in charts, very useful, but I am afraid this is not what CT is about here? I have been wondering whether there is any calculus in the CT which will enable us to derive something new and non-trivial from something simple.
I think I still can’t find the motivation. Maybe just not today.
Raymond
Another message I have got from the lectures is that, they are regarding CT as a useful framework for explaining NN. I have come across a book that tries to explain many different topics in terms of bayesian networks, and topics include Neural Network, Naive Bayes and many more. The book has over 90 chapters and each chapter is a different topic. I have a feeling that, what the CT series is trying to do is something similar which is to convince us that we can understand deep learning topics under the framework of CT. Similarly, the book is trying to demonstrate that all those topics can be understood as bayesian networks.
Checkout benedekrozemberczki on github (benedekrozemberczki (benedekrozemberczki) / Repositories · GitHub)
This person has been working on GNN since some long time. Recommend checking out his work,
Looks interesting, thanks for sharing!
Thanks a lot, Raymond, very helpful resources.
That’s a very good resource book!
I’m still beginning my learning of Category Theory myself and how it relates to GNNs so I’m not sure how to defend it. It is very abstract.
Though I think you might enjoy Taco Cohen’s talk more than Petar’s given your interests.
Got an email today from Stanford that contained a link to : Machine Learning with Graphs | Course | Stanford Online
Might be of interest to the folks on this thread
1 Like