ORFE Professors Boris Hanin, Jason Klusowski and Bartolomeo Stellato each received a grant from the School of Engineering and Applied Science to support their research. 

Professors Hanin and Klusowski received SEAS Innovation Grants, which are intended to foster their research and encourage interdisciplinary collaborations with SEAS faculty but also with colleagues from other academic domains and collaborators from industry.    

Professor Stellato received a Seed Grant, which supports the initiation of a large multidisciplinary collaboration that enhance the visibility and impact of the School of Engineering and Applied Science.

Professor Hanin’s research is at the intersection of theory and practice of deep learning. Specifically, on the theory side, Professor Hanin and his postdoc develop new approaches for analyzing the behavior of equivariant neural networks (ENNs). This analysis is expected to yield new design principles for aligning ENN architecture, optimizer and training data for more reliable and computationally inexpensive training. These principles will then be put to the test, with the goal of obtaining new state-of-the-art results from tasks in particle physics, cosmology, and chemistry. 

Professor Klusowski’s research concerns model selection in regression analysis.  Engineers often face the challenge of selecting the best single model from a range of possibilities that vary in quality.  Traditionally, this selection is based on criteria evaluating a single model’s goodness-of-fit and complexity, or its performance in predicting new data, assessed through cross-validation techniques. Professor Klusowski will explore on which types of data the combined strengths of these possible models have better predictive accuracy than the best single model among them. His research will focus on developing algorithms and theoretical frameworks that connect to constrained optimization and shrinkage in statistics. The ultimate goal is to provide engineers with practical insights into when and by how much a combined model can outperform traditional single-model approaches.

Professor Stellato’s project, entitled “Unsupervised Conditional Generative Machine Learning for Global Nonlinear Optimal Control: Applications in Spaceflight” focuses on advanced onboard autonomous trajectory re-optimization for ambitious future spaceflight missions. This seed grant will bring together Professor Ryne Beeson (MAE), Professor Bartolomeo Stellato (ORFE) and Professor Adji Bousso Dieng (COS), to establish a collaboration with Dr. Donald Ellison from Johns Hopkins University Applied Physics Laboratory (APL), on a new research paradigm of global optimization using end-to-end learning.