Efficient construction of optimal designs for stochastic process models
Richard Boys (Newcastle University, UK)
This talk looks at a new method to determine Bayes optimal (sampling) designs in a fast and computationally efficient way. The method recycles utility calculations and is embarrassingly parallel. It samples designs (and then calculates their utility) according to a particle scheme in which weights are proportional to the current estimated expected utility. The method outperforms many existing methods.