Hyrax: An Extensible Framework for Rapid ML Experimentation and Unsupervised Discovery in the Era of Rubin, Roman, and Euclid
Aritra Ghosh (Airt/r o eghaSh), Drew Oldag, Michael Tauraso, Andrew J. Connolly, Peter Ferguson, Derek Jones, Gourav Khullar, Argyro Sasli, Samarth Venkatesh, Gracia Wang, Maxine West, Dylan Berry, Neven Caplar, Colin Orion Chandler, Tanawan Chatchadanoraset, Michael W. Coughlin, Melissa DeLucchi, Alexandra Junell, Diego Miura, Felipe Fontinele Nunes, Wilson Beebe, Doug Branton, Sandro Campos, Liam Cunningham, Mi Dai, Jeremy Kubica, Konstantin Malanchev, Rachel Mandelbaum, Sean McGuire, Imad Pasha, Dan S. Taranu, Tianqing Zhang · 2026
Read Hyrax: An Extensible Framework for Rapid ML Experimentation and Unsupervised Discovery in the Era of Rubin, Roman, and Euclid by Aritra Ghosh (Airt/r o eghaSh), Drew Oldag, Michael Tauraso, Andrew J. Connolly, Peter Ferguson, Derek Jones, Gourav Khullar, Argyro Sasli, Samarth Venkatesh, Gracia Wang, Maxine West, Dylan Berry, Neven Caplar, Colin Orion Chandler, Tanawan Chatchadanoraset, Michael W. Coughlin, Melissa DeLucchi, Alexandra Junell, Diego Miura, Felipe Fontinele Nunes, Wilson Beebe, Doug Branton, Sandro Campos, Liam Cunningham, Mi Dai, Jeremy Kubica, Konstantin Malanchev, Rachel Mandelbaum, Sean McGuire, Imad Pasha, Dan S. Taranu, Tianqing Zhang (2026) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2605-18959.txt.
Introduction
Aritra Ghosh and thirty-one colleagues present Hyrax, an open-source modular Python framework for machine learning in astronomy. Surveys from Rubin, Roman and Euclid will flood teams with imaging, spectra and time-domain data, shifting the block from model design to infrastructure. Hyrax covers gathering, training, inference and comparison, with multimodal datasets, vector databases for similarity search and latent-space exploration. Five applications on real survey data range from unsupervised discovery of merger candidates among four hundred thousand galaxies to transient classification, lens hunting and dwarf detection.
Catalog record
| CREATOR | Aritra Ghosh (Airt/r o eghaSh), Drew Oldag, Michael Tauraso, Andrew J. Connolly, Peter Ferguson, Derek Jones, Gourav Khullar, Argyro Sasli, Samarth Venkatesh, Gracia Wang, Maxine West, Dylan Berry, Neven Caplar, Colin Orion Chandler, Tanawan Chatchadanoraset, Michael W. Coughlin, Melissa DeLucchi, Alexandra Junell, Diego Miura, Felipe Fontinele Nunes, Wilson Beebe, Doug Branton, Sandro Campos, Liam Cunningham, Mi Dai, Jeremy Kubica, Konstantin Malanchev, Rachel Mandelbaum, Sean McGuire, Imad Pasha, Dan S. Taranu, Tianqing Zhang |
|---|---|
| YEAR | 2026 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 94254 |
| LICENSE | http://creativecommons.org/licenses/by-sa/4.0/ |
| RIGHTS | CC BY SA |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2605.18959
Browse the full library · About the Tower of Babel
Full text
The opening of the work, from the library's hosted text:
Hyrax: An Extensible Framework for Rapid ML Experimentation and Unsupervised Discovery in the Era of Rubin, Roman, and Euclid Facilities: Rubin:Simonyi, Rubin:USDAC, Euclid, Blanco, PO:1.2m, Subaru Numpy ( 41 ) , Scipy ( 108 ) , Astropy ( 10 ; 8 ; 9 ) , Pandas ( 65 ) , Matplotlib ( 47 ) , LINCC Frameworks Python Project Template ( 73 ) , PyTorch ( 6 ) , TorchVision ( https://github.com/pytorch/vision ), PyTorch Ignite ( https://github.com/pytorch/ignite ), MLflow ( 115 ) , Optuna ( 5 ) , umap-learn ( 64 ) , HuggingFace Datasets ( 56 ) , schwimmbad ( 78 ) , pooch ( 103 ) , tqdm ( 24 ) , HoloViews ( 94 ) , Cython ( 14 ) , LanceDB/Lance ( 75 ) , TensorBoard ( https://github.com/tensorflow/tensorboard ), ChromaDB ( https://github.com/chroma-core/chroma ), Qdrant ( https://github.com/qdrant/qdrant ), ONNX ( https://github.com/onnx/onnx ), ONNX Runtime ( https://github.com/microsoft/onnxruntime ), ONNXScript ( https://github.com/microsoft/onnxscript ), Plotly ( https://plotly.com ), PyArrow/Apache Arrow ( https://arrow.apache.org ), Bokeh ( https://bokeh.org ), Datashader ( https://datashader.org ), nested-pandas ( https://github.com/lincc-frameworks/nested-pandas ), Pydantic ( https://docs.pydantic.dev ), psutil ( https://github.com/giampaolo/psutil ), colorama ( https://github.com/tartley/colorama ), more-itertools ( https://github.com/more-itertools/more-itertools ) Aritra Ghosh (Airt/r o eghaSh), Drew Oldag, Michael Tauraso, Andrew J. Connolly, Peter Ferguson, Derek Jones, Gourav Khullar, Argyro Sasli, Samarth Venkatesh, Gracia Wang, Maxine West, Dylan Berry, Neven Caplar, Colin Orion Chandler, Tanawan Chatchadanoraset, Michael W. Coughlin, Melissa DeLucchi, Alexandra Junell, Diego Miura, Felipe Fontinele Nunes, Wilson Beebe, Doug Branton, Sandro Campos, Liam Cunningham, Mi Dai, Jeremy Kubica, Konstantin Malanchev, Rachel Mandelbaum, Sean McGuire, Imad Pasha, Dan S. Taranu, Tianqing Zhang Abstract The NSF-DOE Vera C. Rubin Observatory, Roman Space Telescope, Euclid, and other next-generation surveys will deliver imaging, spectroscopic, and time-domain data at scales that increasingly shift the bottleneck in astronomical machine learning (ML) projects from model design to infrastructure. We present Hyrax, an open-source, modular, GPU-enabled Python framework that supports the full ML lifecycle in astronomy: from data acquisition and training to inference and experiment comparison, with capabilities including multimodal dataset support, integrated vector databases for similarity search, and interactive two- and three-dimensional latent-space exploration for unsupervised discovery. We demonstrate Hyrax’s versatility through five representative applications on real survey data: (i) unsupervised representation learning on ∼ 4 × 10 5 {\sim}4\times 10^{5} Rubin Legacy Survey of Space and Time (LSST) Data Preview 1 (DP1) galaxies, surfacing new merger and low-surface-brightness candidates missing from reference Euclid and Dark Energy Survey catalogs, while also isolating imaging artifacts—all without labeled training data; (ii) hybrid density-based clustering for identifying cluster-scale gravitational lens candidates in DP1 data; (iii) multimodal early-time transient classification in the Zwicky Transient Facility leveraging light curves, spectra, images, and metadata; (iv) supervised false-positive filtering in shift-and-stack searches for distant solar system objects in the Dark Energy Camera Ecliptic Exploration Project survey; and (v) supervised detection of semi-resolved dwarf galaxies in Hyper Suprime-Cam and LSST-like imaging using synthetic source injection. Together, these results demonstrate that Hyrax provides astronomy-specific ML infrastructure that enables systematic discovery and rapid methodological iteration across next-generation astronomical surveys. I Introduction The forthcoming decade in astronomy will feature an unprecedented array of large-scale surveys spanning imaging and spectroscopy, whose depth, breadth, and cadence promise to revolutionize our understanding of the Universe. Among the wide-field imaging surveys, this new generation includes the NSF-DOE Vera C. Rubin Observatory’s Legacy Survey of Space & Time ( 49 , Rubin-LSST;) , the Nancy Grace Roman Space Telescope ( 93 ; 4 , Roman;) , the Euclid mission ( 55 ; 80 ) , and the Chinese Space Station Survey Telescope ( 23 , CSST;) , which will map billions of galaxies across cosmic time at multiple wavelengths. In fact, as shown in Figure 1 , their grasp and sensitivity make them the most information-rich surveys to date, promising transformative scientific returns while simultaneously demanding novel approaches capable of handling the scale and complexity of these datasets. Figure 1: Survey grasp—defined as survey area scaled by angular resolution—is plotted against flux sensitivity for a selection of representative astronomical imaging surveys. Legacy surveys are shown as blue circles, while upcoming wide-fast-deep surveys (e.g., Rubin-LSST, Euclid, Roman, and CSST) are shown as red squares. The diagonal arrow indicates the direction of increasing information content, which scales with both depth and area. This next generation of ground- and space-based surveys will be among the most information-rich to date, enabling data-driven discovery across a broad range of astrophysical domains. Surveys shown are the Cosmic Assembly Near-infrared Deep Extragalactic Legacy Survey ( 40 ; 54 , CANDELS;) , Dark Energy Survey ( 1 , DES;) , Hubble Ultra Deep Field ( 13 , HUDF;) , Galaxy Evolution Explorer Medium Imaging Survey ( 63 , GALEX MIS;) , CSST ( 23 ) , Roman High-Latitude Time-Domain Survey ( 85 , HLTDS;) , Roman High-Latitude Wide-Area Survey ( 93 , HLWAS;) , Rubin LSST ( 49 ) , Sloan Digital Sky Survey ( 114 , SDSS;) , and the Wide-field Infrared Survey Explorer ( 112 , WISE;) . To prepare for and analyze such datasets, machine learning (ML) has been increasingly employed by astronomers for a wide variety of tasks, with applications spanning galaxy evolution, solar system science, transients, anomaly detection, and exoplanets (e.g., 25 ; 45 ; 44 ; 90 ; 29 ; 37 ; 110 ; 96 ; 107 ; 60 ; 36 ; 68 ; 100 ; 22 ; 58 , ; see 46 for a review) . Reflecting this breadth of application, the number of astronomy publications invoking ML has risen dramatically over the past decade, as shown in Figure 2 . However, most of these efforts have remained project-specific, with each group developing bespoke pipelines that duplicate effort and are difficult to extend or adapt beyond their original scientific context. This fragmentation hampers reuse, slows progress, and makes it challenging to scale methods to the demands of upcoming surveys. Designed to address this gap, Hyrax is a modular, flexible, and extensible GPU-enabled framework that provides reusable infrastructure for ML projects in astronomy. Figure 2: Annual number of peer-reviewed astronomy journal articles indexed in the NASA Astrophysics Data System (ADS) from 2010–2024 containing the indicated keywords in either the title or abstract. The steep, sustained rise in publications referencing “machine learning” and “neural network” over the past decade reflects the rapid adoption of these methods across the astronomical community, in contrast to the relatively stable publication rates for evolved sub-domains within astronomy—tracked using the keywords “galaxies”, “cosmology”, and “supernova”. This accelerating interest underscores both the opportunities and the infrastructure challenges that motivated the development of Hyrax. ML in astronomy is often framed primarily in terms of model development and evaluation, but in practice, these tasks represent only a small fraction of the overall effort ( 88 ; 12 ) . The majority of time is typically spent on surrounding tasks: acquiring and organizing data, defining consistent experimental configurations, tracking model outputs, and ensuring that results are reproducible and comparable across iterations. Figure 3 illustrates this reality: while the ML model itself occupies the central red box, it is surrounded by a wide array of non-model components that are routinely required for a complete ML workflow in astronomy. Astronomers need end-to-end frameworks that support