The Department of Math and Computer Science welcomes Dr. Zerotti Woods from the University of Georgia.
Deep Neural Networks have shown much success in solving problems in a diverse set of applications (i.e. computer vision, computational biology, finance, etc). Although we have proof about universal approximation of these networks the problem of training them is known to be very difficult. The ill conditioning of the hessian has been shown to be one of the sources of this difficulty. In this talk we will discuss problems and neural network architecture that causes a ill conditioned hessian. I will also discuss how this can interplay with analysis of high frequency telemetry data taken from a malaria infection on Non Human Primates.
Join us outside Jepson 212 at 4 pm for refreshments before the talk.