Systems Seminar
Importance Sampling Techniques in Monte Carlo Simulation
Prof. Jim Bucklew
UW ECE Dept.
Abstract
Large and/or nonlinear stochastic systems, due to analytic intractability,
must often be simulated in order to obtain estimates of the key performance
parameters. Typical situations of interest could be a buffer overload in a
queuing network or an error event in a digital communication system.
In many system designs or analyses an event of rare probability
is a key parameter of the system's efficacy. To estimate
such a parameter by a brute force direct simulation would require that a
impractically
large number of independent random numbers be generated from the computer's
random number generator. In recent years a variance reduction technique
called {\em importance sampling} has been shown to offer orders of
magnitude improvement over direct Monte Carlo in many common and
important
simulation problems.
In this talk we present an introduction to importance sampling techniques
and some new research results.
Time and Place: Wed., Sept. 13, 3:30-4:30 pm in 4610 Engr. Hall.
SYSTEMS SEMINAR WEB PAGE:
http://www.cae.wisc.edu/~gubner/seminar/