Disaster Planning, Response, and Recovery

Every year, hurricanes threaten coastal areas in the United States, with potentially dramatic consequences on human lives and infrastructures. Currently, decision makers must rely on in-house expertise and do not have the resources to manage large-scale, multi-infrastructure scenarios, and maintain a level of response consistency in different areas of the country. This project studies the computational foundations for disaster planning and last-mile recovery for interdependent infrastructures, including the power and transportation networks, the water supply. Our research has shown that advanced optimization methods, possibly combined with simulation, may provide significant improvements over existing practice. Our research on the water supply has been deployed to aid federal organization in the United States. This project is in collaboration with Los Alamos National Laboratories.

Comet: An Hybrid Optimization System

Comet is an object-oriented programming language for hybrid optimization. providing a smooth integration of mathematical programming, constraint programming, and local search. It features a rich language for programming search procedures, native and transparent parallelism, and declarative visualization. Its constraint-programming solver contains state-of-the-art algorithms covering a wide variety of constraints, including scheduling and rostering constraints, soft constraints, and set constraints. Comet also introduced constraint-based local search to apply local search to high-level declarative models through the concept of invariants. Its rich search and control language also allow for deep integration of constraint and mathematical programming, contraint programming and local search, large neighborhood search, and many other hybridizations.  A commercial version of Comet, free for academic and research purposes, is available from Dynadec, following a technology transfer from Brown University. This project is in collaboration with the university of Connecticut.

Online Stochastic Combinatorial Optimization

Online decision making under uncertainty and time constraints represents one of the most challenging problems for robust intelligent agents. In an increasingly dynamic, interconnected, and real-time world, intelligent systems must adapt dynamically to uncertainties, update existing plans to accommodate new requests and events, and produce high-quality decisions under severe time constraints. This project studies new paradigms and algorithms to approach these problems by combining techniques from artificial intelligence, operations research, and theoretical computer science. Successful applications of online stochastic optimization include real-time vehicle routing and dispatching, ambulance dispatching, kidney exchanges, the scheduling of pharmaceutical projects, and real-time reservation systems.

Global Optimization

Global optimization aims at finding all solutions to systems of nonlinear equations or, more generally, to find global optima to nonlinear, non convex  problems. The research aims at combining techniques from artificial intelligence, numerical analysis, and mathematical programming to enclose and prune the set of solutions. Techniques for the reliable enclosures of differentiable equations are also studied.  

Intelligent Real-Time Control of Smart Energy Markets

This project aims at designing and implementing intelligent and robust planning and control mechanisms that balance thermal, wind, and solar energy with electricity storage and demand response in dynamically priced electricity markets with significant stochasticity in front and behind the meter. It focuses on addressing the issues raised by (1) the increasingly fundamental role of demand response, (2) the increased stochasticity on the demand and supply side, and (3) the increasingly complex nature of the forecasting and optimization problems necessary to plan and control the grid. The scientific contributions include novel forecasting and optimization algorithms to automate demand response and incorporate renewable energies in a robust and economically viable smart grid. This project is in collaboration with the University of Connecticut.

Testing and Verification

This project investigates the use of constraint programming and constraint-based local search for automatic test generation and formal verification of hardware and software. Examples include the automatic generation of architectural tests for processor validation and the verification of partial correctness of Java programs from a high-level specification.  This project is in collaboration with the University of Nice and was supported by Intel.

Computational Biology

This project looks at a variety of computational problems in computational biology, including protein structure prediction, RNA structural segmentation, folding pathways between RNA Secondary Structures, and various alignment problems at different levels of abstraction. This project is in collaboration with Boston College.