Invited Talk
in
Workshop: 3rd Workshop on High-dimensional Learning Dynamics (HiLD)
Nathan Srebro (TTIC& University of Chicago), Is A Good Input Distribution All You Need?
Nati Srebro
Abstract:
What functions representable by neural nets are tractably learnable? Complexity results tell us that not all of them are, and we have been on a quest to understand what is the subclass of functions that are learnable. In this talk I will revisit and question this view, putting an emphasis on the input distribution, rather than the target function, and arguing that perhaps all functions are easy, it’s just a matter of the input distribution. This also leads to understanding much of the current success of deep learning in terms of “Positive Distribution Shift”.
Chat is not available.
Successful Page Load