Geometric approaches to parameter estimation for differential equations.
- Date: 03/26/2008
Joe Tien (Fred Hutchinson Cancer Research Center)
University of British Columbia
Parameter estimation for differential equations is a fundamental
problem, but work in this area is incomplete. One difficulty is the
presence of local minima which trap optimization algorithms, resulting
in poor fits. Incorporating geometric features of the differential
equations into the optimization algorithms can alleviate this problem.
These structural features are combined with numerical improvements via
automatic differentiation, and applied to neural models and data
related to breathing.
This work has resulted in my becoming interested in statistical
inference. If time permits, I will mention some HIV data I have been
working with on viral strains which have escaped immune system
MITACS Math Biology Seminar 2007