This project focuses on the principled design, enhancement, analysis and application of model-directed hybrids, which combine model-directed optimization and traditional optimization techniques. The key idea is to use models of the problem landscape constructed by model-directed optimization techniques to facilitate decisions about the nature and likely effectiveness of particular local search procedures and appropriate neighborhood structures for those procedures. The importance of the developed techniques will be demonstrated in a broad spectrum of applications, including problems in computational physics, biology, chemistry and operations research.
The project will have transformative effects on computational optimization mainly because it will automate the design of advanced neighborhood structures and problem-specific operators applicable to broad classes of optimization problems, including problems with noise, dynamic landscape, complex structure and multiple conflicting objectives. Besides advancing computational optimization, the project will have a strong impact on a variety of other disciplines in science, engineering and commerce because it will provide practitioners in these disciplines with tools that allow practical solutions of problems intractable with current techniques. Furthermore, the project will provide methodology for development of advanced techniques for landscape analysis as well as theoretical study of advanced hybrids of metaheuristics and traditional optimization techniques.
To complement the research goals, the project puts a strong emphasis on broader impacts with the focus on advising student researchers, facilitating graduate mentoring of undergraduates, participating in multidisciplinary and collaborative projects, ensuring effective dissemination of research outcomes, organizing student competitions, boosting the local research infrastructure, and supporting groups underrepresented in science and engineering.
