The Mathematical and Resource Optimization program of the Office of Naval Research (ONR) recently awarded two grants to Lehigh Industrial and Systems Engineering (ISE) faculty for a total over $1M.
Lehigh ISE Professor Frank E. Curtis and Associate Professor Daniel P. Robinson were awarded a three-year $621K grant from ONR for their project entitled “Stochastic Algorithms for Data-Driven Constrained Continuous Optimization.” The project will contribute to cutting-edge research in the field of optimization methods for machine learning and data science applications. In particular, the project focuses on optimization algorithms for informed learning, where prior information about a system or process can be incorporated in the learning process rather than rely on purely data-driven techniques. Curtis and Robinson will leverage their renowned expertise on constrained optimization algorithms.
Curtis says: “We are extremely grateful to ONR for the support through this grant, which will enable us to significantly advance the role of state-of-the-art mathematical optimization techniques in modern data science applications.” Robinson affirms: “Machine learning and artificial intelligence will continue to play an increasing role in defense operations, such as for motion tracking systems and the operation of autonomous vehicles. We’re thankful for this support from ONR for our work, which will provide foundational advances for long-term impact.”
Lehigh ISE Timothy J. Wilmott Endowed Chair Professor and Department Chair Luis Nunes Vicente was awarded a three-year ONR $542K grant as PI, titled “New Sampling and Descent Paradigms for Stochastic Black-Box Optimization”. This project aims to develop state-of-the-art stochastic methodologies and rigorous theoretical analyses to expand the knowledge of derivative-free optimization in several innovative directions. A new tail bound condition for function estimation is introduced which allows for a substantial improvement in the required sample sizes. Another key contribution is the combination of sequential and adaptive sampling in each iteration of an algorithm, which allows the algorithm to make significant progress before the adaptive budgets are exhausted.
Nunes Vicente says: I am grateful to ONR to support research in such an exciting topic. Derivative-free optimization finds practical application in many defense problems involving control and learning. Practical settings involve expensive function simulations where efficiently adapting the sample sizes becomes crucial.