ExoMiner

<div style="text-align: center;"> <img src="/others/images/exominer_logo.png" width="250" height="250" alt="Exominer Logo"> </div>

Written by Miguel Martinho (miguel.martinho@nasa.gov)

Introduction

This project's mission is to develop, test, and deploy automated machine learning-based methods to sift ('mine') through transit photometry data from exoplanet survey missions such as Kepler and TESS and inform subject matter experts (SMEs) on potential transiting planet candidates.

Current main goals

The main goals of the ExoMiner project are:

1. Perform classification of transit signals in Kepler and TESS data;
2. Create vetted catalogs of Threshold Crossing Events (TCEs) for Kepler and TESS runs for the exoplanet community.
3. Validate new exoplanets using Kepler and TESS data.

ExoMiner Pipeline [NEW]

The ExoMiner Pipeline is a fully integrated pipeline from TIC IDs to ExoMiner prediction scores for the corresponding TESS SPOC TCEs in 2-min/FFI data. This pipeline makes use of Podman as a container framework to provide a seamless experience to the user - no need for setting up the code repository and install package dependencies! Simply get the Podman image for your system's architecture using the manifest. See the documentation in here to get started.

For those interested in running the pipeline without resorting to Podman, see Running the pipeline without Podman.

Repository Overview

The repository consists of the following main blocks:

1. Data preprocessing: preprocess data products (e.g. light curve FITS files) to generate a catalog of transit signal features to be used for training and evaluating models, and to run inference on it. Code under src_preprocessing.
2. Model training/evaluation/prediction: train and evaluate models on the preprocessed data. Run inference on the data using trained models. Code under src.
3. Model evaluation using cross-validation under src_cv.
4. Hyperparameter Optimization: use code under src_hpo to run hyperparameter optimization experiments to find an optimized architecture for a given task.
5. Model Development: use code under models to access and modify ExoMiner architectures.

Source Data

All data used in this project are publicly available. Generally, the data used consist of:

• TCE and Objects of Interest (e.g. KOI and TOI catalogs) tables available in archives/respositories such as NExSci, ExoFOP, TEV and MAST;
• Light curve and target pixel FITS files created by the Kepler/TESS Missions, and data products generated by the Kepler/TESS Science Processing Operations Center (SPOC) available in archives such as the MAST.

Models

Models currently implemented in models:

1. ExoMiner++ [CURRENT]: Improved architecture for TESS. Used in TESS paper (see References).

References

For more detailed information see the following publications:

• ExoMiner for Kepler "ExoMiner: A Highly Accurate and Explainable Deep Learning Classifier that Validates 301 New Exoplanets", published 2022 February 17 in the Astrophysical Journal, Volume 926, Number 2.
• ExoMiner with Multiplicity Boost for Kepler "Multiplicity Boost of Transit Signal Classifiers: Validation of 69 New Exoplanets using the Multiplicity Boost of ExoMiner", published 2023 June 26 in the Astronomical Journal, Volume 166, Number 1.
• ExoMiner for TESS 2-min "ExoMiner++: Enhanced Transit Classification and a New Vetting Catalog for 2-Minute TESS Data", published 2025 October 27 in the Astronomical Journal, Volume 170, Number 5
  • Data repository on Zenodo: here
  • Interactive vetting catalog: here

Credits

This work was developed by members of the Data Sciences Group, DASH, Intelligent Systems Division (Code-TI) at NASA Ames Research Center (NASA ARC).

• Main Contributors

  • Hamed Valizadegan^1,2 (PI), hamed.valizadegan@nasa.gov
  • Miguel Martinho^1,2 (Co-I), miguel.martinho@nasa.gov

• Collaborators

  • Active Collaborators

    • Doug Caldwell^1,3
    • Jon Jenkins^1,3
    • Joseph Twicken^1,3

  • Past Collaborators

    • Stephen Bryson¹
    • Jeff Smith^1,3

• Developers

  • Active Developers

  • Past Developers

    • Andrés Carranza ^2,5 (Unfolded phase time series for transit signal classification)
    • Fellipe Marcellino² (Transit detection using Kepler data)
    • Jennifer Andersson⁴ (Kepler to TESS transfer learning)
    • Kaylie Hausknecht^2,6 (Explainability framework using Kepler data)
    • Laurent Wilkens² (Kepler)
    • Martin Koeling⁴ (Kepler to TESS transfer learning)
    • Nikash Walia² (Kepler)
    • Noa Lubin ⁴ (Kepler)
    • Pedro Gerum⁴ (Kepler, Kepler non-TCE classification)
    • Patrick Maynard^2,5,7 (Kepler to TESS transfer learning)
    • Sam Donald⁴ (Kepler to TESS transfer learning)
    • Theng Yang^2,7 (Label noise detection in Kepler data)
    • Hongbo Wei^2,6,7 (Kepler non-TCE classification, KOI classification, Kepler to TESS transfer learning)
    • Stuti Agarwal ⁶ (Difference image)
    • Joshua Belofsky ^2,5 (Difference image)
    • Charles Yates ^2,5 (Unfolded phase time series for transit signal classification, Kepler to TESS transfer learning)
    • William Zhong ⁵ (Difference Image)
    • Ashley Raigosa⁷ (TESS SPOC FFI)
    • Saiswaroop Thammineni⁷ (Transit Encoding)
    • Kunal Malhotra⁷ (Transit Detection)
    • Eric Liang⁷ (Transit Encoding)
    • Ujjawal Prasad⁸ (Transit Detection)
    • Adithya Giri⁷ (Brown Dwarfs vs Planets Classification; Structured and Adversarial Training for Transit Classification Robustness)
    • Josue Ochoa⁷ (Transit Detection)
    • Aniket Mittal⁷ (XAI for ExoMiner)

1 - NASA Ames Research Center (NASA ARC)
2 - KBR Inc.
3 - The SETI Institute
4 - NASA International Internship Program (NASA I²)
5 - NASA Internships, Fellowships & Scholarships (NIFS)
6 - Volunteer Internship Program (VIP)
7 - NASA Office of STEM Engagement (OSTEM)
8 - NASA-Chabot High School Learning Experience (NASA-CHSLE)

Acknowledgements

We would like to acknowledge people that in some way supported our efforts:

• A big thank you to all the interns who dedicated their time to supporting the development of ExoMiner!
• The SPOC team (Jon Jenkins, Joe Twicken, Doug Caldwell) at NASA Ames/SETI for their insight, knowledge, and support!
• David Armstrong for an insightful discussion that improved our work.
• Megan Ansdell for providing information on their code and work on Exonet.
• The main contributors (HV and MM) are supported through TESS XRP 2022 contract 22-XRP22 2-0173, NASA Academic Services Mission (NAMS) contract number NNA16BD14C as well as the Intelligent Systems Research and Development 3 (ISRDS-3) Contract 80ARC020D0010.
• Resources supporting this work were provided by the NASA High-End Computing (HEC) Program through the NASA Advanced Supercomputing (NAS) Division at Ames Research Center.
• This work has made use of data products created by the Kepler/TESS Science Processing Operations Center (SPOC) pipeline at NASA Ames Research Center.
• This work has made use of data from the European Space Agency (ESA) mission Gaia (https://www.cosmos.esa.int/gaia), processed by the Gaia Data Processing and Analysis Consortium (DPAC, https://www.cosmos.esa.int/web/gaia/dpac/consortium). Funding for the DPAC has been provided by national institutions, in particular the institutions participating in the Gaia Multilateral Agreement.
• This work includes data collected by the TESS mission. Funding for the TESS mission is provided by the NASA's Science Mission Directorate.
• This research is based on