Sep 4, 2018 · 1 min read
I have a question about the paragraph about James Gleick’s book on Chaos. While attempts at analyzing deep nets have begun using more and more sophisticated mathematics (out of necessity, of course!), do you think we’ll really have to go to the extent of invoking Chaos theory to explain, for example the miraculous generalization powers such networks seem to possess?
I guess you’re thinking about some sort of stochastic approximation technique that will eventually require the analysis of a highly nonlinear limiting ODE that could be chaotic (?).
