Daniel Muthukrishna
Research Scientist, MIT Kavli Institute for Astrophysics and Space ResearchAstroAI Fellow, Center for Astrophysics | Harvard & Smithsonian
I build AI that separates the physics of the universe from the instruments we use to observe it, and I deploy it on NASA space missions and the world’s largest sky surveys.
Scientific AI for data-intensive physics
Causal, uncertainty-aware and simulation-aligned methods built for the realities of physical instruments and data.
Causal foundation models
Representations that disentangle the physics of a source from the properties of the instrument that measured it, so that models generalise across telescopes, sensors and scientific domains. Demonstrated on variable stars and on multi-survey galaxy imaging.
Generative models of instrument systematics
BEAM learns and forecasts scattered light in NASA TESS images without paired training data. Its blind forecast was used to plan TESS’s January 2026 observations of the interstellar comet 3I/ATLAS.
Real-time discovery in sky surveys
RAPID, DASH and classifier-based anomaly detection for the transient sky, with simulation pretraining that matches fully supervised performance using an order of magnitude fewer labels. Deployed live on the Lasair broker for Rubin LSST and ZTF alerts.
Exoplanet discovery with deep learning
I lead the Astronet group for NASA’s TESS mission, a twelve-person team of engineers and scientists delivering the production neural-network pipeline that vets planet candidates and has contributed to hundreds of discoveries.
Background
I am a Research Scientist at the MIT Kavli Institute for Astrophysics and Space Research and an AstroAI Fellow at the Center for Astrophysics | Harvard & Smithsonian. My research develops scientific AI for data-intensive physics: causal foundation models that separate the physics of a source from the instrument that measured it, generative models that learn instrument systematics from the data themselves, and real-time classification, anomaly detection and decision tools for wide-field surveys. My work has been published at NeurIPS, ICML and ICLR and in the leading astrophysics journals.
At MIT I lead the Astronet group for NASA’s TESS mission, coordinating twelve software engineers, research scientists, faculty and PhD students (including volunteers from Google and Meta) who deliver the production deep learning pipeline for vetting exoplanet candidates. At the Center for Astrophysics I coordinate the multi-modal AI group and organise the AstroAI seminar series. My open-source tools RAPID and DASH are used by surveys around the world, and my classifiers run live on the UK-led Lasair alert broker for the Vera C. Rubin Observatory’s LSST and for ZTF.
I completed my PhD in Astrophysics at the University of Cambridge in 2021, working on real-time classification and anomaly detection for time-domain surveys. Before that I earned first-class honours degrees in Physics and in Electrical and Aerospace Engineering at the University of Queensland and worked as a radio-frequency engineer designing microwave systems. I lecture the course Data-driven Astronomy: Machine Learning and Statistics for Modern Astronomy, and I have supervised fourteen students from high school to PhD whose work has appeared at NeurIPS, ICML, MNRAS and RASTI.
Recent highlights
Three papers accepted at NeurIPS 2026: BEAM (Tufto & Muthukrishna) and Learning What’s Real (Mercader-Perez et al.) in the main track, and the PhyTS benchmark for scientific time series in the Evaluations and Datasets track.
The open textbook Deep Learning for Astrophysics from the NASA Cosmic Origins AI/ML Science and Technology Interest Group is published in Research Notes of the AAS, including my chapter and lecture on recurrent neural networks.
Invited talk at CERN, Gravitational Waves in the Era of AI, Geneva.
Delivered the NASA AI/ML Science and Technology lecture on recurrent neural networks for astronomical light curves. Watch the recording.
NASA’s TESS re-observed the interstellar comet 3I/ATLAS; the 28-hour movie of the comet I compiled from the data was featured by NASA and BBC Sky at Night Magazine.
Appointed Research Scientist at the MIT Kavli Institute and AstroAI Fellow at the Center for Astrophysics | Harvard & Smithsonian.
Started two new projects: a NASA TESS Guest Investigator programme (PI) to correct scattered light in TESS images with generative AI, and an MIT Generative AI Impact Consortium collaboration with industry partners to build a time-series foundation model for astrophysics.
Selected papers
-
Reversing the Roles of Signal and Noise: Modeling Noise in Astronomical Images without Paired Data
NeurIPS 2026, main track
CodeBEAM -
NeurIPS 2026, main track; ICLR 2026 Workshop on Foundation Models for Science
-
Causal Foundation Models: Disentangling Physics from Instrument Properties
ICML 2025 Workshop on Foundation Models for Structured Data
-
Real-time Detection of Anomalies in Large-scale Transient Surveys
MNRAS, 517, 393 (2022)
-
RAPID: Early Classification of Explosive Transients Using Deep Learning
PASP, 131, 118002 (2019)
† marks students under my direct supervision.
Courses and supervision
- Designed and lecture the thirteen-week course Data-driven Astronomy: Machine Learning and Statistics for Modern Astronomy (Cambridge Centre for International Research, 2023–present).
- NASA Cosmic Origins AI/ML Science and Technology lecture on recurrent neural networks (2026), now a chapter of the open textbook Deep Learning for Astrophysics.
- Organised and lectured the University of Cambridge’s intensive summer computing course for mathematics students (2018, 2019).
- Fourteen students supervised to date; alumni have gone on to Stanford, Caltech, UT Austin, OpenAI, Tesla, Waymo, Verizon and Capital One.
Panels, refereeing and collaborations
- Review panellist for the National Science Foundation (2024, 2026), NASA’s TESS Guest Investigator Program (2024, 2026) and Schmidt Sciences’ LINCC Frameworks mid-project review (2025).
- Referee for ApJ, Nature Astronomy, A&A, MNRAS, AJ, IEEE journals and the NeurIPS and ICML machine learning for the physical sciences workshops.
- Member of the LSST Dark Energy Science Collaboration (co-developer of the PLAsTiCC challenge), the Young Supernova Experiment and the Habitable Worlds Observatory AI/ML Working Group.
- Co-organiser of the Boston Astrophysics x Machine Learning Hackathon (2024) and the Accelerating Physics with ML workshop at MIT (2023); local organising committee, TESS Science Conference III.