Some approaches and techniques for efficient control of experiments

Classical design of experiments, however, does not take into account the fact that prior variances of model parameters do not coincide with their posterior variances. The estimated variance of the model parameters and the predicted response depends not only on the design but also on the observed values of the output variable. Our research has shown that optimal plans obtained by taking these posterior variances into account perform better than those derived using purely classical methods. Therefore, it is important to develop algorithms for constructing optimal designs sequentially that take into account these posterior variances. The sequential nature of experiment is ignored in the classical framework. Specifically, the goals of the proposed project are: develop algorithms for constructing optimal plans taking into account the sequential nature of experimentation, develop software for generating optimal designs, demonstrate the usefulness of the methodology and the software in real life examples.