September 10, 2014

Speaker: Dr. Peter Klages, Dunlap Institute for Astronomy and Astrophysics, University of Toronto

Title: Computational Imaging from the Microscopic to the Megaparsec

Abstract: With the advent of general purpose graphics processing unit (GPU) computing, and the recent boom in commodity GPUs, the speed at which we can process and analyze data has increased by magnitudes for a certain subclass of problems. In this seminar Dr. Klages will speak about two separate projects he has worked on that utilize these new technologies to achieve things that were not possible a decade ago.

The first project deals with the microscopic realm, and uses computers to reconstruct 3D volumes from interference patterns of visible light recorded with a Complementary Metal Oxide Silicon (CMOS) sensor (digital inline holographic microscopy), and the second project deals with cosmologically important scales, recording radio waves at discrete sites and computing complex visibilities and formed beams from the data (Canadian Hydrogen Intensity Mapping Experiment). Dr. Klages will discuss some of the background physics that are part of these two very different, yet related, topics, and will talk about how this translates to algorithms to be used on central processing units (CPUs) or GPUs, noting the gains in performance that can be had when the specific architectures are kept in mind.