Abstract:

In this talk, I will talk about three myths widely spread in deep learning community. (1) Expressiveness: To represent a function, deep structure is much more efficient than the shallow ones. (2) Optimization: Gradient descent works well even though deep learning is a non-convex optimization problem. (3) Generalization: Deep networks generalize well.

Registration site:   https://goo.gl/erHU7Y

Seminar homepage: https://goo.gl/iPKJEq