I mentioned a while ago the notion of a very basic neural network being the product of data dependent matrix choices. Eg. A₄B₃B₂A₁x for data dependent choices of A or B.
I kinda extended the idea here.
https://archive.org/details/data-dependent-operator-selection-layers
I’m trying not to over share on the internet. It can get rather unpleasant for the person sharing. Especially on larger groups like reddit. And with zero up-side for the person sharing.
Why do it then? Is the question I ask myself.
1 Like