Many engineering and scientific projects require planning experimental activities in order to optimize an objective. Such planning involves jointly reasoning about both the budget-limited resource constraints among activities along with the utility of potential information that they may provide. Unfortunately, for many real-world planning problems, with rich structure among potential activities, tools from classic experimental design are not directly applicable due to their simplifying assumptions and poor scalability. This project aims to transform the practice of experimental planning by developing new algorithms that account for the complexities exhibited in a wide range of domains.
The project involves three key activities. First, the experimental design description language (EDDL) is being created for formally modeling complex real-world experimental domains. Second, novel planning algorithms are being developed for efficiently computing high-quality solutions to problems expressed in EDDL. Third, work with bioengineers is assessing and improving the usability of the tools and producing benchmark problems based on real and simulated bioengineering data. The creation of the language and benchmarks will help facilitate algorithm comparisons for continued progress by the wider research community.
The broader impact of the project is to facilitate experiment planning in a wide range of experimental domains for which there are currently no available computational tools. Currently, in such domains, planning is largely ad-hoc and often done without computer support. Our research has broad economic impact potential by helping engineers and scientists to get the most value out of limited experimental resources. The project also advances high school, undergraduate, and graduate education in the areas of computer science and bio-engineering, with an emphasis on recruiting female students. The students will get the unique experience of working across disciplines and research labs.
