University of Houston

The goal of this PhD is to provide algorithms and a computational framework for complex optimization problems of objective functions that are computationaly intensive; that are stochastic with large noise ratio; and that exhibit multimodal landascapes. In particular, one needs to extract all the relevant clusters around competitive extrema within the noise interval. Such problems are frequent in modeling biological or ecological systems. Ultimately, our solution should use a cost-effective computing environment, such as Volunteer Computing and low-cost HPC systems, in order to be accessible to almost every concerned scientist.
We chose the Virtual Prairie (ViP) project as our testbed, since it presents multiple optimum design problems having the above properties. The ViP project aims at understanding the dynamics and achieving optimum design of prairies. Indeed, the interest in prairie is increasing, since it was recently proved that they are a good source of biofuel and a soil-decontamination agent.
We propose a framework that uses Genetic Algorithms enhanced with Niching Techniques in order to satisfy the multi-optima optimization objective. Parallel versions of these algorithms would reduce the execution time necessary for optimizations of real-world scientific applications. These parallel algorithms will be deployed on a hybrid computing platform composed of a fat node or possibly an HPC system, used to evolve the population, and of thousands of Internet-connected compute nodes, provided by volunteer computing and enabled by the Berkeley Open Infrastructure for Network Computing, to benefit from the embarassing level of parallelism of the fitness function evaluations. Optimum design problems from the ViP project will be used for testing and validating the suggested framework.