When
Where
Speaker: Brian Toner, Program in Applied Mathematics
Title: Intro to the mathematics of MRI
Abstract: Magnetic resonance imaging, an imaging modality unique in its high contrast of soft tissue, relies on the physics of nuclear spins of Hydrogen atoms to create signals that can be measured in k-space. In this talk, we will cover (briefly) some of the fundamentals of MR physics and touch on some of the image reconstruction techniques that have both emerged and sometimes been supplanted by newer techniques over the 50 year history of MRI.
Speaker: Edward McDugald, Program in Applied Mathematics
Title: Intro to the diffusion models in generative AI
Abstract: Diffusion models have emerged as a powerful tool to sample data distributions from noise, and are useful for generating ensembles, performing super resolution, and solving sparse sensing problems. In this talk we will focus on score-based diffusion models, where a neural net is trained to learn the time dependent score function of the forward process, which is then used to sample from the data distribution using a reverse SDE. We aim to derive the loss function in a clear and concise fashion, as well as lay out the training and sampling procedures.