Communications and Signal Processing Seminar
Not Enough Measurements, Too Many Measurements
Michael McCannPostdoctoral Research AssociateComputational Mathematics, Science and Engineering (CMSE), Michigan State University
WHERE:
Remote/Virtual
WHEN:
Thursday, October 22, 2020 @ 4:00 pm - 5:00 pm
This event is free and open to the publicAdd to Google Calendar
This event is free and open to the publicAdd to Google Calendar
SHARE:
Abstract: One focus of image reconstruction research is recovering a high-quality image from a small number of its noisy measurements. At the same time that we struggle due to this lack of data, we also have access to a bounty of images—e.g., over 14 million natural images in ImageNet—and measurements—e.g., almost 300 CT scans in the Patient CT Projection Data Library and thousands of knee and brain MRI scans in the fastMRI Dataset. In this talk, I will discuss several projects on the theme of using “too many” measurements to help us reconstruct from “not enough” measurements: using a database of full-dose CT scans to aid in low-dose reconstruction; using generative modeling to reconstruct from numerous (but noisy) projections in cryo-EM single particle analysis; and designing regularizers for inverse problems directly from example images.
Bio: Michael McCann is a postdoc with Saiprasad Ravishankar’s group in the Dept. of Computational Mathematics, Science and Engineering at Michigan State University. Mike’s research focus is improving biomedical image reconstruction algorithms using tools from signal processing, optimization, and machine learning. He was previously with Michael Unser’s group at EPFL, Lausanne, Switzerland, where he worked on straight-ray tomography reconstruction. Before that, he did his PhD in Jelena Kovačević’s group (Carnegie Mellon University, Pittsburgh, Pennsylvania) on histology image analysis. Even before that, he worked on EEG-based brain-computer interfaces with Jane Huggins at the University of Michigan, Ann Arbor, Michigan.
Join Zoom Meeting https://umich.zoom.us/j/97598571292
Meeting ID: 975 9857 1292
Passcode: XXXXXX (Will be sent via email to attendees)