Unless stated otherwise, colloquia are scheduled for Thursdays 4:15-5:15pm in WH-100E with refreshments served from 4:00-4:15 pm in WH-102.
Friday May 6, 3:45-4:45pm, WH-100E (NOTE SPECIAL DATE AND TIME)
Speaker: Minghao W. Rostami (Syracuse University)
Topic: Biofluid as a big data challenge
The simulation of a fluid around dynamic biological structures,
such as bacteria and cilia, entails solving large systems of Partial
Differential Equations (PDEs) and Ordinary Differential Equations (ODEs).
This boils down to working with large-scale, dense matrices with very few
zero entries. We first show that these matrices are “data sparse”, that
is, the large amount of data stored in them can be significantly
compressed. We then present fast algorithms for matrix-vector
multiplication and linear solves involving these matrices. Our methods do
not require constructing the large dense matrices and can achieve huge
savings in storage and time. We also show how parallel-in-time methods can
further speed up the simulation of a biofluid when spatial parallelization
“saturates”. In addition, a data-driven, reduced order model discovered by
deep learning will be discussed. It allows us to describe the movement of
a fluid without using complex PDEs.