friendly_aixi

3 ms·

1. 1.▲
   by friendly_aixi 6y ago

   1) It's definitely not the first. Other methods have universal guarantees of some form or other with well quantified rates of convergence, e.g. k-NN would be the best known example. 2) There are some restrictions on the class of densit
2. 2.▲
   by friendly_aixi 6y ago

   1) Indeed, GLNs don't learn features... but I would claim they do learn some notion of an intermediate representation, it's just different from the DL mainstream -- in particular its closely related to the inverse Radon transform
3. 3.▲
   by friendly_aixi 6y ago

   The result here is stronger, in the sense that typical NN universality results are statements with respect to just capacity (and not how you optimise them). Here, the result holds with respect to both capacity + a choice of suitable no regr
4. 4.▲
   by friendly_aixi 6y ago

   Convolution is a linear operation; in the case of images, you can view it as a multiplication with a doubly block circulant matrix. I can't see any barriers to hybrid approaches here, though it seems difficult to avoid using backpropag