[drumnerd]

I think about cones all the time when doing machine learning. If I have an object O that can be mapped to A and B I can learn a function from A to B or B to A if I can generate Os. That’s my view of self supervised learning.