A central theme in algorithm design is that of making decisions in the presence of uncertainty. In contrast to the usual setting where all the information is available up-front, often the input is not known precisely when we have to make the decisions, but is revealed along the way. We face such decision-making situations all the time, e.g., when deciding on driving routes in the presence of traffic, or dividing our time between chores that take uncertain amounts of time. This project aims to design algorithms for a wide class of such problems using only predictions about the future input. Research on decision-making under uncertainty started nearly seventy years ago, but it has gained much momentum in recent years. This is partly due to numerous applications in scheduling, transportation, electronic commerce etc., and partly because of the vast amounts of data that allow us to make good predictions about the future. This project will model a collection of fundamental and practically relevant problems in this area, and develop techniques to obtain algorithms with provable guarantees on their performance. Research results from this project can bring together communities in computer science with those in operations research, stochastic control and machine learning. The educational component of this project includes the engagement of graduate and undergraduate students in research, and the development of a new graduate course in this subject.
The project will model predictions about the uncertain input using probabilistic models. This approach, called stochastic optimization, is one of the most widely-used approaches to model uncertainty. This project aims to make progress on basic problems in scheduling, path-planning and routing, packing and covering, and submodular maximization, in this stochastic setting. The focus of this project will be on two kinds of algorithms for optimization in these settings: non-adaptive algorithms (where all decisions are made in one shot), and adaptive ones (where the decisions are made incrementally, based on random outcomes observed along the way). One of the goals is to broaden the scope of investigation further by considering settings where the underlying random quantities exhibit correlations; this is in contrast to the usual assumptions of independence.
This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.
