The Atacama Cosmology Telescope: Machine Learning Driven Tools for Detecting Millimeter Sources in Timestream Pre-processing
Simran K. Nerval, Erika Hornecker, Yilun Guan, Zeling Zhang, Adam Hincks, Emily Biermann, J. Richard Bond, Justin Clancy, Rolando Dünner, Allen Foster, Carlos Hervías-Caimapo, Renée Hložek, Thomas W. Morris, Sigurd Næss, John Orlowski-Scherer, Cristóbal Sifón, Jesse Treu · 2025
Read The Atacama Cosmology Telescope: Machine Learning Driven Tools for Detecting Millimeter Sources in Timestream Pre-processing by Simran K. Nerval, Erika Hornecker, Yilun Guan, Zeling Zhang, Adam Hincks, Emily Biermann, J. Richard Bond, Justin Clancy, Rolando Dünner, Allen Foster, Carlos Hervías-Caimapo, Renée Hložek, Thomas W. Morris, Sigurd Næss, John Orlowski-Scherer, Cristóbal Sifón, Jesse Treu (2025) 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-2503-10798.txt.
Introduction
Simran Nerval and co-authors present machine learning tools that clean raw data from a sky survey telescope, in this 2025 paper. The Atacama Cosmology Telescope watches faint microwave light from the early universe. Its detectors record long streams of data spiked with brief glitches. Old cleaning rules cut every large spike alike, whether it came from noise, a cosmic ray, or a real sky source. That bluntness can erase real signals.
The new pipeline sorts short features in the raw streams before the main cleaning steps. A learned classifier tells likely sky sources from junk spikes. Genuine point sources can then be kept and handled with care. The tools work alongside the older rules, adding judgment where fixed thresholds had none.
Catalog record
| CREATOR | Simran K. Nerval, Erika Hornecker, Yilun Guan, Zeling Zhang, Adam Hincks, Emily Biermann, J. Richard Bond, Justin Clancy, Rolando Dünner, Allen Foster, Carlos Hervías-Caimapo, Renée Hložek, Thomas W. Morris, Sigurd Næss, John Orlowski-Scherer, Cristóbal Sifón, Jesse Treu |
|---|---|
| YEAR | 2025 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 77 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2503-10798
Browse the full library · About the Tower of Babel
Full text
The opening of the work, from the library's hosted text:
The Atacama Cosmology Telescope: Machine Learning Driven Tools for Detecting Millimeter Sources in Timestream Pre-processing Simran K. Nerval, Erika Hornecker, Yilun Guan, Zeling Zhang, Adam Hincks, Emily Biermann, J. Richard Bond, Justin Clancy, Rolando Dünner, Allen Foster, Carlos Hervías-Caimapo, Renée Hložek, Thomas W. Morris, Sigurd Næss, John Orlowski-Scherer, Cristóbal Sifón, Jesse Treu Abstract We present a new pipeline utilizing machine learning for classifying short-duration features in raw time-ordered data (TOD) of cosmic microwave background survey observations. The pipeline, specifically designed for the Atacama Cosmology Telescope, works in conjunction with the previous TOD preprocessing techniques that employ statistical thresholding to indiscriminately remove all large spikes in the data, whether they are due to noise features, cosmic rays, or true astrophysical sources, in a process called “data cuts”. This has the undesirable effect of excising real astrophysical sources, including transients, from the data. The classification pipeline demonstrated in this work uses the output from these data cuts and is able to differentiate between electronic noise, cosmic rays, and point sources, enabling the removal of undesired signals while retaining true astrophysical signals during TOD preprocessing. We achieve an overall accuracy of 90% in categorizing data spikes of different origin and, importantly, 94% for identifying those caused by astrophysical sources. Our pipeline also measures the amplitude of any detected source seen more than once and produces a subminute-to-minute light curve, providing information on its short timescale variability. This automated pipeline for source detection and amplitude estimation will be particularly useful for upcoming surveys with large data volumes, such as the Simons Observatory. Astropy 5 , 6 , 7 , cutslib , enact , enlib , Matplotlib 29 , moby2 , modAL 18 , NumPy 23 , pandas 33 , 44 , pickle , pixell , scikit-learn 40 , 14 , SciPy 46 , Seaborn 47 https://github.com/amaurea/enact/ https://github.com/ACTCollaboration/enlib https://github.com/ACTCollaboration/moby2/ https://github.com/simonsobs/pixell I Introduction Ground-based microwave astronomy has experienced a revolution in the past two decades, thanks to the deployment of large aperture telescopes that have thousands of detectors, increased their resolution, and advances in detector and read-out technology that have drastically improved their sensitivity while also allowing for simultaneous observation of multiple wavelength bands. For example, the Atacama Cosmology Telescope (ACT; Swetz et al. Super , Thornton et al. Super , Henderson et al. Super ) observed the cosmic microwave background (CMB) with \sim 3000 detectors across up to three frequency bands, covering six bands from {\sim} 30 to 270 GHz during its operation from 2007 to 2022. 1 1 1 Jupyter notebooks outlining how to use the latest ACT data products (ACT DR6) can be seen at https://github.com/ACTCollaboration/DR6_Notebooks , including a glitch classification notebook outlining the methods discussed in this paper. ACT will be succeeded by the Simons Observatory (SO), which is initially set to deploy over 60,000 detectors split between large- and small-aperture telescopes across six frequency bands from {\sim} 30 to 280 GHz 3 , 2 . During observations, detectors, such as transition edge sensors, record the incident power at a cadence of a few hundred hertz as the telescope scans in azimuth at a constant elevation. In the ACT survey, the data timestreams of all detectors are packaged into files known as “time-ordered data” (TOD), each containing roughly 11 minutes worth of data. Using knowledge of the telescope pointing, which is recorded during data collection, the TODs are later processed into microwave sky maps using maximum likelihood map-making 19 , 17 , 4 . In addition to using the CMB directly as a probe of cosmological physics, an emerging use of CMB experiments is the study of the time-varying microwave sky. Wide-area surveys contain thousands of radio sources; the majority of them are blazars that emit microwave light that changes in flux over days/weeks. The South Pole Telescope (SPT) collaboration 15 published a pilot study of one blazar (PKS 2326-502) observed in the microwave bands 28 . Surveys like SO will provide much longer baselines of observations of such objects: the field of mm-transient studies is expected to grow significantly in the coming decade. Similarly in the realm of planetary science, ACT and SPT have also characterized the millimeter flux of asteroids 16 , 39 , and ACT constrained the existence of a “Planet 9” in our Solar System 36 . Finally, ACT and SPT have successfully measured dozens of transients in their data that are mainly flaring stars, with the majority consisting of short-duration (minutes- to hours-long) stellar flares 48 , 37 , 31 , 22 , 43 , 12 . Additionally, as part of a blind transient search, ACT has detected classical nova YZ Ret during its outburst, making it only the second millimeter observation of this kind 12 . ACT also published a targeted search for extragalactic transients that included gamma ray bursts (GRBs), tidal disruption events (TDEs), and supernovae (SN), placing upper-limits on the microwave flux of the target objects 26 . More sensitive data from the upcoming SPT, SO, and CMB-S4 1 surveys will produce detections in the millimeter of extragalactic transients, specifically GRBs, which will provide valuable information on the emission processes of these powerful explosions 20 , 2 . CMB surveys are designed to measure the power of faint microwave radiation imprinted on all angular scales across the sky. However, the data collected by ground-based CMB telescope like ACT are dominated by microwave emission from water vapor in the atmosphere and are additionally contaminated from, e.g., instrumental noise and cosmic rays colliding with the detectors. While instrumental and atmospheric noise can be modeled in the microwave map-making process to produce unbiased, maximum likelihood maps 19 , the mapmaker is specifically vulnerable to biases from short, high-amplitude, non-Gaussian data bursts in the TOD, and therefore depends on these bursts being excised from the TOD before map-making. Such transient events can be caused by sporadic pathologies from the electronic readout system, cosmic rays, electromagnetic interference, radio frequency interference modulated by telescope motion, ionizing radiation, digitization artifacts, etc 34 , 8 , 9 . Cosmic rays, for example, sometimes produce rapid spikes in the TODs of single detectors and may sometimes also warm up a region of the detector polarization array (PA) near the impact, thereby inducing a signal spike in multiple detectors (see 34 for a study on cosmic rays for Planck High Frequency Instrument). In this paper, we refer to all of these transient phenomena as “glitches”. Bright astrophysical point sources can also cause high-amplitude, non-Gaussian spikes in the TOD and may be mistaken for a glitch. This may be avoided if the coordinate of the source is known before pre-processing. However, this also means that any point source or astrophysical transient that is not already known in advance of map-making can be removed from the data set entirely. We note that while the specific phenomena we describe here are commonly encountered in ACT timestreams, other transient sources may manifest differently in different telescopes depending on their readout schemes and observing conditions. It is therefore imperative when preprocessing TODs from CMB telescopes to excise as many glitches as possible while still retaining as many true astrophysical transients as feasible. As the volume of data observed by current and future experiments increases, traditional techniques that rely on