Þorsteinn Jónsson (02/07/18)

Anders Claesson, June 29, 2018

Math Colloquium

Speaker: Þorsteinn Jónsson, University of Guelph

Title: Variational Techniques for Learning Distributions of Data

Location: V-147 (VR-II)
Time: Monday 2 July at 10:30


In this talk I will give an overview of a set of techniques that allow us to define latent variable models of data generating distributions.
To this end we introduce neural networks as an efficient and a surprisingly effective way of finding variational parameters that best fit some specified objective.
I will discuss a couple of different approaches that can be taken to define this objective and show you some interesting results.