I don't see anything here beyond "neural networks generalize linear operators." (But of course they're more interesting for nonlinear applications) I think neu

I don't see anything here beyond "neural networks generalize linear operators." (But of course they're more interesting for nonlinear applications) I think neutral nets are such a generalized thing that tons of stuff can be shown to map to them. I saw a talk once about evolution being equivalent to neural networks: consider a genome as a set of neuron weights, and your iterative refinement as new generations in evolution. Maybe there's something deep there, but maybe if you make your model arbitrarily complicated enough you can make it look like any system you want. Which is the whole idea with neural nets.