The research objective of this award is to create new simulation optimization algorithms that combine rigorous theoretical performance guarantees with the robust empirical behavior of a class of random search techniques called the model-based methods. The research approach is based on integrating the principle of the well-known Expectation-Maximization (EM) algorithm from the field of statistics into model-based methods. In particular, through exploiting a novel connection to the EM algorithm, this research will investigate a unifying framework to design and implement new model-based algorithms for solving a broad class of simulation optimization problems with very modest computational effort. These algorithms will be studied in terms of their properties (such as convergence and convergence rate) using a fusion of theories and tools from EM, stochastic approximation, and Quasi-Newton methods. A variety of applications from biostatistics to electric power systems will also be tested for the purposes of evaluating the practical utility of the developed techniques and algorithms.
If successful, the resulting techniques will have applicability in a wide array of industry and science sectors. Through collaboration with bio-statisticians, the developed algorithms will be applied to optimal drug dose-response experimental designs, with potential benefits to health care. In addition, the intended applications to electric power systems will also promote synergy among different disciplines. The research resulting from this project will be disseminated through publications, software development, and participation at national and international conferences. This award will also be closely integrated with the education and training of students in mathematical science and engineering by incorporating new developments into the advanced courses taught by investigators at different institutions, and promoting the participation of female students in research.
