Neural Networks for Algebraists
Speaker:
Sara Jamshidi Zelenberg
Illinois Institute of Technology
Visiting Assistant Professor of Applied Mathematics
Description:
Neural networks are often treated as black-box methods for solving unknown problems. In this talk, however, we will attempt to make them a little more transparent, especially from an algebraic perspective. We'll discuss some of the considerations when attempting to use these tools in the field of algebra. We will also discuss what precisely we are doing when we use these tools; specifically what are we constructing from a geometric perspective when we use ReLU activations. The talk will primarily focus on feed-forward setups, but, time permitting, some discussion will be given to other setups, such as recursive neural networks. (Note: this talk is adapted from one given an MPI.)
Event Topic
Nonlinear Algebra and Statistics (NLASTATS)