Non-Stationary Gaussian Processes at High-Performance Computing Scales for the Space Domain
Amanda Muyskens | 22-ERD-028
Executive Summary
We will develop novel machine-learning algorithms with principled uncertainty quantification for large, complex scientific data that support high-fidelity modeling at heretofore intractable scales. This capability will enable scientific and security advancements in the space domain, empowering data-intensive missions with scalable, predictive tools featuring statistical interpretability and uncertainty quantification.
Publications, Presentations, and Patents
Imène R. Goumiri,. Amanda L. Muyskens, Benjamin W. Priest, and Robert E. Armstrong, “Light Curve Forecasting and Anomaly Detection Using Scalable, Anisotropic, and Heteroscedastic Gaussian Process Models” (Conference, Advanced Maui Optical and Space Surveillance Technologies (AMOS) Conference proceedings, 2023).
R. Bidese, C. Eleh, Y. Zhang, R. Molinari, N. Billor, B. W. Priest, A. L. Muyskens, “Stellar Blend Image Classification Using Computationally Efficient Gaussian Processes (MuyGPs)” (Presentation, Symposium on Data Science and Statistics (SDSS) Extended Abstract Proceedings, 2023).
Muyskens, Amanda L., Imène R. Goumiri, Benjamin W. Priest, Michael D. Schneider, Robert E. Armstrong, Jason Bernstein, and Ryan Dana. “Star-Galaxy Image Separation with Computationally Efficient Gaussian Process Classification.” The Astronomical Journal 163, (04 2022): 148. https://doi.org/10.3847/1538-3881/ac4e93.
Imène Goumiri, “Light Curve Forecasting and Anomaly Detection Using Scalable, Anisotropic, and Heteroscedastic Gaussian Process Models” (Presentation. Advanced Maui Optical and Space Surveillance Technologies AMOS Conference, Maui, HI, September 19-22, 2023).
Alex Geringer-Sameth, “Scalable Gaussian processes for big-data challenges in astronomy” (Presentation at LLNL’s Physical Life Sciences External Review Committee Meeting held at Lawrence Livermore National Laboratory (LLNL), Livermore, CA, 2023).
Min Priest, “Space Domain Awareness and Astronomy: Scale, fidelity, and new frontiers” (Presentation at LLNL’s Computing External Review Committee Meeting, Lawrence Livermore National Laboratory, Livermore, CA, 2023).
Rafael Bidese, “Stellar Blend Image Classification Using Computationally Efficient Gaussian Processes (MuyGPs)” (Presentation at Symposium on Data Science and Statistics (SDSS) held in St. Louis, MO, May 23-26, 2023).
Alec Dunton,“Fast Gaussian Process Prediction via Local Cross-Validation and Precomputation” (Poster Presentation at Conference on Data Analysis (CODA), Santa Fe, NM, March 7-9, 2023).
Imene Goumiri, “Gaussian Process Regression Algorithm (MuyGPs) Science Applications” (Poster Presentation, Conference on Data Analysis (CODA), Santa Fe, NM, March 7-9, 2023).
Alec Dunton,“Scalable Gaussian Process Hyperparameter Optimization via Coverage Regularization” (Poster Presentation at NeurIPS Workshop on Gaussian Processes, Spatiotemporal Modeling, and Decision-making Systems, New Orleans, LA, Dec 2, 2022).
Amanda Muyskens,“Understanding the Universe with Applied Statistics” (Presentation for LLNL’s Data Science Institute (DSI) Spotlight, Livermore, CA, November 17, 2022). https://www.youtube.com/watch?v=FLMkpm210qs.
Min Priest, “MuyGPs: A Whirlwind Tour of Scalable Gaussian Process Applications” (Presentation, Dark Energy Science Collaboration Meeting, Jan 2023).
Min Priest, “Light Curve Completion and Forecasting using Fast and Scalable Gaussian Processes (MuyGPs)” (Presentation at the Space Systems Command (SSC)'s SDA Data Science Working Group (DSWG) meeting, Jan, 2023).
Min Priest, “HPC-Scale Statistical Modeling with Astronomy Applications” (Presentation, Sustainable Research Pathways (SRP) Project Matching Meeting, 2023).
Imène Goumiri, “Light curve completion and forecasting using fast and scalable Gaussian processes (MuyGPs)” (Presentation, Advanced Maui Optical and Space Surveillance Technologies (AMOS) Conference, Maui, HI, September 27-30, 2022).
Amanda Muyskens,“MuyGPs: Scalable Gaussian Process Model Estimation with Uncertainty Quantification” (Presentation, American Statical Association’s (ASA) Section on Defense and National Security (SDNS) Webinar Series, Livermore, CA, September 21, 2022).
Amanda Muyskens,“MuyGPs: Scalable Gaussian Process Model Estimation with Uncertainty Quantification” (Presentation, Quality and Productivity Research Conference (QPRC), San Francisco, CA, June 13-16, 2022).
Amanda Muyskens,“MuyGPs: Scalable Gaussian Process Model Estimation with Uncertainty Quantification” (Presentation, Space Systems Command (SSC)'s SDA Data Science Working Group (DSWG) meeting, Lawrence Livermore National Laboratory, Livermore, CA, June 7-8, 2022).
Amanda Muyskens,“MuyGPs: Scalable Gaussian Process Model Estimation with Uncertainty Quantification" (Presentation, DMS Statistics and Data Science Seminar Series, Auburn University, Auburn, AL, March 17, 2022).