Vic Anand
Current Research
For complex anatomical structures such as the hand, it is necessary to explicitly distinguish between model structure (i.e., the preconceived morphology) and parameter values (i.e., the particulars of that structure). The inevitable discrepancies between predicted and measured data can be attributed to unsatisfactory parameter values, inadequate model structure or both. Today’s biomechanical models consist of manually assembled structures where only the parameter values are systematically adjusted to explain and/or reproduce experimental data. Thus, improving current models necessitates that we explicitly investigate how the assumed model structure fundamentally determines and limits model behavior.
I am developing unsupervised algorithms that simultaneously infer both the model structure and parameter values of hidden systems. A genetic algorithm (GA) evolves models of the target system. The GA simultaneously varies the topology and parameters of its individuals; the only assumption made by the GA is about the composition of the set of “building blocks” used to assemble model topologies. Once models have been evolved based on data previously observed from system, tests are evolved. The fitness of a test is based on its ability to produce disagreement among the models; we hypothesize that a test with high fitness will extract more information from the hidden system than a test with low fitness and thereby reduce the total number of tests on the hidden system. Once a good test is found, it is performed on the hidden system. The new data obtained from the hidden system is then used to evolve a new generation of models.
This method seeks the best combination of model structure and corresponding parameters that best explain the data. It also seeks to reduce the amount of testing performed on the system being modeled. In this way, we hope that models can begin to clarify how disease and treatment affect the type, connectivity, properties, parameters and interactions of available “building blocks” such as bones, tissues, tendons, muscles, neural circuits, etc.
Background
C.V.
Publications
Valero-Cuevas FJ, Lipson H, Santos VJ, and Anand V. Shifting to population-based models and inferring model structure from data are two directions that will enhance the clinical usefulness of modeling. XXth Congress of the International Society of Biomechanics and 29th Annual Meeting of the American Society of Biomechanics, Cleveland, OH, MC 201 ISB Technical Group: Simulation Symposium, August 1, 2005, p.963.
Anand V., Lipson H., and Valero Cuevas F.J. (2005) Blind Inference of Nonlinear Cable Network Topology from Sparse Data. Proceedings of the 2005 Genetic and Evolutionary Computation Conference, June 2005, Washington D.C., USA. Late breaking paper.
Spears, W. and Anand, V. (1991) A Study of Crossover Operators in Genetic Programming. Sixth International Symposium of Methodologies for Intelligent Systems.
Education
Ph.D. Mechanical Engineering, Cornell University, in progress
MBA Finance, Carnegie Mellon University, 2000
S.B. Mechanical Engineering, Massachusetts Institute of Technology, 1995
Work Experience
|
Employer |
Location |
Dates |
Title |
|
Science Applications International Corporation (SAIC) |
McLean, VA |
2001 – 2003 |
Senior Business Analyst |
|
Deloitte Consulting |
Detroit, MI |
2000 – 2001 |
Senior Consultant |
|
Ford Motor Company |
Dearborn, MI |
1997 – 1998 |
Financial Analyst |
|
Ford Motor Company |
Livonia, MI |
1995 – 1997 |
Manufacturing Engineer |
Internships
|
Employer |
Location |
Dates |
Nature of Work |
|
Deloitte Consulting |
Detroit, MI |
Summer 1999 |
Management consulting |
|
General Motors Corporation |
Warren, MI |
Summer 1994 |
Internal combustion engine research |
|
MIT |
Cambridge, MA |
1994 – 1995 |
Lab TA for a C programming class (1.00) |
|
MIT |
Cambridge, MA |
1993 – 1994 |
UROP at the Sloan Automotive Lab; engine test cell setup, machining, etc. |
|
U.S. Naval Research Lab |
Washington, DC |
Summers 1989, 1990 |
Genetic algorithm research |
Personal Interests
Running
I have been running since junior high school. My favorite race distance is the 10-miler (16K). I ran the Army Ten Miler in both 2002 and 2002. I have also run the Pittsburgh Marathon and the Detroit Marathon, and scores of 5K’s and 10K’s.
Bicycling
I have done three bicycle tours – three weeks in the Canadian Rockies and Montana, one week along the coast of Lake Michigan, and one week in the badlands of North Dakota. One of these years, when I have the time, I’ll train for and do some bike races.