Department of Mathematics and Statistics
|Thursday, April 13, 2022
|1:15pm – 2:15pm
|Wei Yang, Binghamton University
|Unitary Neural Networks
Vanishing / exploding gradient problem often arises in training deep neural networks. One way to make the neural network more stable is to replace the weight matrices Unitary / Orthogonal matrices. But this comes with the cost of a more expensive training cost. We will study some applications and methods of training in the literature.