Name: Kevin Brown
Title: Assistant Professor
Department: Biomedical Engineering
Personal/Lab Website: http://kbrown.research.uconn.edu
Program/Department Website: http://www.bme.uconn.edu/faculty-staff/core-faculty/21474-2
Office Telephone: (860) 486-6975
Keywords | Categories: Bayesian and nonparametric statistics, Biomedical Engineering, Chemical and Biomolecular Engineering, complex systems, inverse problems, mathematical modeling, networks, systems biology
Multimodal Data Fusion: Combining multiple data sets with complementary spatial and temporal resolution in order to obtain an integrated view of a process of interest is a difficult problem that arises in many disparate contexts. Two examples are (i) combining satellite measurements with ground-based sensors in earth science, and (ii) fusing simultaneous electroencephalographic and functional magnetic resonance measurements of human brain activity. I have developed BICAR, a new algorithm to extract paired sources of interest from two sets of sensor data with vastly different degrees of sensor coverage and sampling rate. BICAR is objective, flexible, extensible, specific, and symmetric. Sloppy Models: Modeling complex systems presents several challenges: (i) uncertain parameters and model structure, (ii) the necessity of using simplified dynamics, and (iii) limited data availability.
I designate problems with these challenges sloppy models. Sloppiness is ubiquitous, and sloppy models are a well-defined class of problem that arises in nonlinear systems with weak parameter constraints, and the resulting model space geometry has implications for understanding these systems. By using ensemble methods, I can generate confidence intervals on model predictions and yield robust, falsifiable models. These kinds of calculations can also be directly extended to both model selection and optimal experimental design.