Raoul LePage , Department of Statistics and Probability,

Michigan State University

Monday, 20 September 2004

4:10 PM

E202, Engineering

MC-INVERSE is a Monte Carlo method able to produce a solution of simultaneous linear equations in any number of positive variables. It is joint work with myself, Krzysztof Podgorski and Michal Ryznar. Our setup includes ill-posed systems of equations and is technically a superset of Bayes' method. MC-INVERSE will be described together with its properties of existence and consistency (recovery of true solution given enough information). Its behavior when presented with inconsistent information will also be described. The Monte Carlo draws samples in solution space by means of an arbitrary working probability measure. Our solution is then obtained from this Monte Carlo by an importance sampling (i.e. weighting) argument.

Refreshments will be served at 3:45 p.m. in Room 008 of the Statistics Building