(Statistics, University of Manitoba)
A practical and efficient Monte Carlo random sampling method for statistical computation and global optimization
|Date||Monday, November 2, 2009|
In engineering and many other fields, scientific research often involves numerical optimization, integration or visualization of multi dimensional functions. When the dimension of the problem is moderate or high, deterministic methods are usually difficult and stochastic methods are more practical. In this talk I will introduce a Monte Carlo method for efficient generation of random points from any multivariate distribution. This method provides a practical tool to explore a high dimensional function and to quickly identify the "significant region" within the domain of the function. When applied to global optimization problems, this method is further developed to a mode-pursuing algorithm which systematically generates more sample points in the neighbourhood of the function modes while statistically covering the entire search space. Examples and engineering applications will be used to demonstrate this method.