The Machine Learning and Instrument Autonomy strives to research, develop, and infuse machine learning and other data science supporting technologies to advance robotic exploration and science.
Recent News
-
▸ February 10th, 2026 by Ryan McGranaghan
Ryan McGranaghan along with the leaders of the JPL Science Understanding from Data Science (SUDS) initiative concluded paper collection for a Special Collection in the Journal of Geophysical Research Machine Learning and Computation based on the JPL SUDS Initiative. The collection is entitled, “Science Understanding from Data Science: Transformative Science Through the Convergence of Data Science and Physical Science.” The collection received 30 total submissions across four JGR journals and represents an important artifact from the SUDS effort. Special collection organizers include Lukas Mandrake, Erika Podest, Amy McGovern (U. Oklahoma), Rajesh Gupta (UCSD), Barbara Thompson (NASA GSFC), Chris Bard (NASA GSFC), Jake Lee, Samuel Berndt, Marcel Kaufman.
-
▸ July 3rd, 2025 by Jake Lee
Jake Lee, Michael Kiper, David Thompson, and Phil Brodrick are coauthors of “SpecTf: Transformers enable data-driven imaging spectroscopy cloud detection”, published in the Proceedings of the National Academy of the Sciences (PNAS). This work presents a new deep learning model for cloud detection in imaging spectroscopy data. SpecTf leverages a spectroscopy-specific transformer architecture to generate high-accuracy cloud masks, significantly outperforming the current EMIT baseline. Notably, SpecTf requires only spectral information, demonstrating strong performance and interpretability through its attention mechanism, revealing physically meaningful spectral features. The model also exhibits potential for cross-instrument generalization. The resulting cloud mask will be deployed as a new EMIT product.
-
▸ June 11th, 2025 by Ryan McGranaghan
Ryan McGranaghan planned, convened, and hosted a first-of-its-kind event on June 9th titled “Boston-Area Complex Risk Science: Exploring new frontiers and a new community for understanding risk.” Held at Northeastern University, it gathered 40 individuals from across the Boston area to identify non-obvious gaps in risk science, particularly around multi-hazard interactions, cascading consequences, and vulnerability, and to explore how complexity science and social science can help us understand and address them. Institutions involved included MIT Lincoln Laboratory, the MIT Media Lab, Harvard, NASA Lifelines, the Electric Power Research Institute, New England Independent System Operator (NE-ISO), Dartmouth University, the Massachusetts Clean Energy Center, among many others.
-
▸ June 11th, 2025 by Mario Damiano
ExoReL (Exoplanet Reflected Light retrieval), a Bayesian inverse retrieval framework used to interpret exoplanetary reflected light spectra, has been open sourced by PI and developer Mario Damiano. ExoReL has been instrumental in supporting research related to the Habitable World Observatory and the Starshade probe, appearing in over eight peer-reviewed journal articles. Researchers will also utilize ExoReL to analyze upcoming data from the Roman Space Telescope, specifically reflected light spectra of cold gaseous giant planets.