Sajant Anand
Sajant Anand
Home
Talks
Publications
Light
Dark
Automatic
Publications
Type
Preprint
Date
2021
On the Power of Shallow Learning
We prove that any NNGP or NTK kernel of any wide, deep fully-connected network architecture can be achieved in just a single hidden layer with a specially-designed activation function. We validate this surprising result with several experiments.
J. B. Simon
,
Sajant Anand
,
M. R. DeWeese
Cite
×