2023 NSF HDR Ecosystem Conference

Keynote Speakers

Yisong Yue

Yisong Yue is a Professor of Computing and Mathematical Sciences at the California Institute of Technology, and a Principal Scientist at Latitude AI. He was previously a research scientist at Disney Research. Before that, he was a postdoctoral researcher in the Machine Learning Department and the iLab at Carnegie Mellon University. He received a Ph.D. from Cornell University and a B.S. from the University of Illinois at Urbana-Champaign. Yisong is also the Senior Program Chair of the ICLR 2024 (International Conference on Learning Representations).

Yisong’s research interests are centered around machine learning, and in particular getting theory to work in practice. To that end, his research agenda spans both fundamental and applied pursuits, from novel learning-theoretic frameworks all the way to deployment in autonomous driving on public roads. His work has been recognized with multiple paper awards and nominations, including in robotics, computer vision, sports analytics, machine learning for health, and information retrieval. At Latitude AI, he works on machine learning approaches to behavior modeling and motion planning for autonomous driving.

Talk Title: AI for Adaptive Experiment Design

Abstract: Experiment design is a hallmark of virtually all research disciplines. In many settings, one important challenge is how to automatically design experiments over large action/design spaces. Furthermore, it is also important for such a procedure to be adaptive, i.e., to adapt to the outcomes of previous experiments. In this talk, I will describe recent progress in using data-driven algorithmic techniques for adaptive experiment design, also known as active learning and Bayesian optimization in the machine learning community.  I will cover a few case studies, including in personalized clinical therapy, nanophotonic structure design, and protein design. Motivated by these applications, I will show how to incorporate real-world considerations such as safety, preference elicitation, and multi-fidelity experiment design into the Bayesian optimization framework, with new algorithms, theoretical guarantees, and empirical validation.

David Hogg

David W. Hogg is Professor of Physics and Data Science at New York University, where he has been since 2001, and Group Leader for Astronomical Data at the Flatiron Institute, where he has been since its founding in 2016. His work spans all scales in astrophysics from the measurement of cosmological parameters to the discovery of extra-solar planets. His research is interdisciplinary between astronomy, applied mathematics, and engineering; it is designed to enable precision measurements in complex data, accurate calibration of instrumentation, efficient operations for observatories, and discoveries of hidden objects or phenomena. He is a co-founder of NYU’s Center for Data Science, and he is the Chair of the Advisory Council of the Sloan Digital Sky Survey.

Talk Title: Machine learning in astrophysics: Open questions and unsolved problems

Abstract: The astronomical literature has blown up with machine learning in the last five years; now a significant fraction of all refereed papers make use of machine learning. The main roles for ML in astronomy are in classification, outlier detection, and emulation of expensive simulations. Many times when ML is introduced into an astronomical project, new risks are also introduced, in part because the epistemology and ontology of ML don’t match the ontology and epistemology of the natural sciences: In ML only the data exist; and in ML correctness is judged only by performance on held-out data. I raise open questions for the future of ML in astronomy related to these risks, and recommend paths for new research. I also show some places where the flexibility and epistemology of ML methods make them the best tools for the job. One example is in causal inferences, where openness to confounding effects is critical. At present, there aren’t many important astronomical discoveries that are directly attributable to ML, but I expect this to change as the community addresses the open questions.