This grant provides funding for the development of new methodologies for improving the efficiency of simulations of large-scale stochastic systems, such as those arising in manufacturing, telecommunications, production/inventory, finance, and transportation. The basic ideas underlying many of the award techniques is to extract or utilize more information from a simulation than is done in the standard approaches.
The methods are related to existing variance-reduction techniques such as control variates, which collect additional data that are used to modify the estimator, leading to less variability. The Principal Investigators propose to develop other approaches that similarly exploit ignored information from simulations to improve the estimator. Several of the proposed techniques take a particular realization of the simulation output and construct many other possible realizations from it. Estimates of the performance measure of interest are computed from each realization, and the estimates are then averaged. This leads to lower variability than
the standard approach.
The Principal Investigators plan to establish the validity of these methodologies under various assumptions and develop computationally efficient implementations of the ideas. One of the desired outcomes is to show that the proposed methods are optimal under certain conditions. If successful, the results of the project will lead to significant improvements in the efficiency of simulations of large-scale systems. The basic ideas underlying the methods are quite general and versatile, and they can be combined with other existing simulation techniques.
|Effective start/end date||6/15/99 → 5/31/04|
- National Science Foundation: $189,406.00