pith. sign in

arxiv: 1907.04566 · v1 · pith:CGGDFJZ4new · submitted 2019-07-10 · 🌌 astro-ph.IM

Prospects for the Use of Photosensor Timing Information with Machine Learning Techniques in Background Rejection

Samuel Spencer , Thomas Armstrong , Jason Watson , Garret Cotter This is my paper

Pith reviewed 2026-05-24 23:46 UTC · model grok-4.3

classification 🌌 astro-ph.IM
keywords machine learningimaging atmospheric Cherenkov telescopeswaveform timingbackground rejectionphotosensor readoutsevent classificationastroparticle physicsCherenkov photon arrival times
0
0 comments X

The pith

Creating seven 2D histograms of waveform parameters lets machine learning use photosensor timing to improve background rejection beyond what integrated charge images alone achieve.

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The paper examines whether timing details in the full readout waveforms from imaging atmospheric Cherenkov telescope cameras can help machine learning separate signal events from background more effectively than the standard integrated-charge images. It proposes feeding the machine learning algorithm a combination of the usual charge image plus seven new two-dimensional histograms that capture timing-related waveform parameters. This combined input is presented as superior to applying the same machine learning methods to the charge image by itself. The authors also note that the same timing-based approach could be examined in other high-speed imaging experiments.

Core claim

A set of seven additional two-dimensional histograms of waveform parameters, when supplied to the machine learning algorithm together with the integrated charge image, constitutes a highly effective way to exploit timing information and yields better event classification performance than using the waveform integrated charge alone.

What carries the argument

Seven 2D histograms of waveform parameters that encode arrival-time information from the photosensor readouts.

If this is right

  • Machine learning classifiers can extract usable timing discrimination directly from existing camera waveforms without new hardware.
  • Current and future IACT cameras that record full waveforms can improve background rejection by including these histograms as additional input channels.
  • The same histogram-based timing encoding can be tested in other experiments that record high-speed photosensor data.
  • Event classification accuracy improves when timing structure is represented explicitly rather than left implicit in the charge image.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • If the method works, analysis pipelines could reach lower energy thresholds or higher sensitivity by rejecting more background at the same signal efficiency.
  • The approach may transfer to any detector that records both spatial and temporal photon arrival data, such as certain neutrino or dark-matter experiments.
  • One could test whether a smaller or differently chosen set of waveform histograms retains most of the gain.

Load-bearing premise

The timing details recorded in the waveforms carry additional information that helps distinguish event types and is not already contained in the integrated charge image.

What would settle it

A side-by-side test in which adding the seven timing histograms produces no gain in classification metrics such as ROC area or background rejection efficiency at fixed signal efficiency would falsify the central claim.

Figures

Figures reproduced from arXiv: 1907.04566 by Garret Cotter, Jason Watson, Samuel Spencer, Thomas Armstrong.

Figure 1
Figure 1. Figure 1: Left: The charge image from a 44TeV proton event. Centre: The associated timing histogram. Right: The first smoothed waveform to survive the peak charge cut in the image. We investigate two scenarios. In the first, we attempt to differentiate between gamma-rays from a point source against a diffuse background of proton and electron events (as would be the case for most observations). In the second, we atte… view at source ↗
Figure 2
Figure 2. Figure 2: Receiver Operator Characteristic (ROC) curves for the four methods described displaying the True Positive Rate (TPR) against the False Positive Rate (FPR). The associated Area Under Curve (AUC) metrics are also shown. In the point source run case, we consider the ROC curves for classification between γ-rays and protons and between γ-rays and electrons. This demonstrates the combined classification power of… view at source ↗
Figure 3
Figure 3. Figure 3: Classification categorical accuracy as a function of energy for the point source run (left) and the diffuse run (right). The width of the x error bars depicts the size of the bin. the timing histogram significantly outperforms Method C from the current literature. In the case of the diffuse events, for all four methods there is significantly more confusion between electron and gamma-ray events. This sugges… view at source ↗
read the original abstract

Recent developments in machine learning (ML) techniques present a promising new analysis method for high-speed imaging in astroparticle physics experiments, for example with imaging atmospheric Cherenkov telescopes (IACTs). In particular, the use of timing information with new machine learning techniques provides a novel method for event classification. Previous work in this field has utilised images of the integrated charge from IACT camera photomultipliers, but the majority of current and upcoming IACT cameras have the capacity to read out the entire photosensor waveform following a trigger. As the arrival times of Cherenkov photons from extensive air showers (EAS) at the camera plane are dependent upon the altitude of their emission, these waveforms contain information useful for IACT event classification. In this work, we investigate the potential for using these waveforms with ML techniques, and find that a highly effective means of utilising their information is to create a set of seven additional two dimensional histograms of waveform parameters to be fed into the machine learning algorithm along with the integrated charge image. This appears to be superior to using only these new ML techniques with the waveform integrated charge alone. We also examine these timing-based ML techniques in the context of other experiments.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

1 major / 0 minor

Summary. The manuscript investigates prospects for using photosensor waveform timing information with machine learning for background rejection in imaging atmospheric Cherenkov telescopes (IACTs). It claims that constructing seven additional two-dimensional histograms from waveform parameters and feeding them to ML algorithms alongside the integrated-charge image yields superior event classification performance compared to using the integrated-charge image alone; the work also places these timing-based techniques in the context of other experiments.

Significance. If the empirical results hold under detailed scrutiny, the proposed method would offer a practical route to exploit full waveform readout (already available in current and upcoming IACT cameras) for improved discrimination between gamma-ray and hadronic showers, potentially increasing sensitivity without requiring new hardware.

major comments (1)
  1. [Abstract] Abstract: the central claim that the seven-histogram approach 'appears to be superior' is stated without any quantitative performance metrics, dataset descriptions, model architectures, or baseline comparisons, rendering it impossible to assess whether the timing information supplies discriminatory power beyond the charge image.

Simulated Author's Rebuttal

1 responses · 0 unresolved

We thank the referee for their review. We address the single major comment below.

read point-by-point responses

  1. Referee: [Abstract] Abstract: the central claim that the seven-histogram approach 'appears to be superior' is stated without any quantitative performance metrics, dataset descriptions, model architectures, or baseline comparisons, rendering it impossible to assess whether the timing information supplies discriminatory power beyond the charge image.

    Authors: We agree that the abstract, as a concise summary, omits the quantitative details (specific AUC improvements, dataset sizes, CNN architectures, and direct baseline comparisons) that appear in Sections 3–4 of the manuscript. These sections report the simulated IACT datasets, the seven 2-D histogram construction, the network training, and the performance gains relative to charge-image-only inputs. To address the concern directly, we will revise the abstract to include the key quantitative metrics demonstrating the improvement. revision: yes

Circularity Check

0 steps flagged

No significant circularity; empirical ML comparison is self-contained

full rationale

The paper reports an empirical study that directly compares ML classification performance on integrated-charge images alone versus the same images augmented by seven timing-derived 2-D histograms. No derivation, uniqueness theorem, or parameter fit is invoked whose output is then relabeled as a prediction. The central claim is tested by standard train/test evaluation on simulated or real data and is therefore falsifiable outside any internal construction. No self-citation supplies a load-bearing premise, and no equation or ansatz reduces to its own inputs by definition.

Axiom & Free-Parameter Ledger

0 free parameters · 0 axioms · 0 invented entities

No free parameters, axioms, or invented entities mentioned in the abstract.

pith-pipeline@v0.9.0 · 5746 in / 890 out tokens · 18014 ms · 2026-05-24T23:46:58.124926+00:00 · methodology

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

17 extracted references · 17 canonical work pages · 6 internal anchors

  1. [1]

    CTA Consortium, Science with the Cherenkov Telescope Array, World Scientific, Singapore, (2019)

    work page 2019

  2. [2]

    Hillas, Cerenkov light images of EAS produced by primary gamma rays and by nuclei, Proc

    A. Hillas, Cerenkov light images of EAS produced by primary gamma rays and by nuclei, Proc. 19th I.C.R.C. (La Jolla) 3 445 (1985)

    work page 1985

  3. [3]

    Application of Deep Learning methods to analysis of Imaging Atmospheric Cherenkov Telescopes data

    I. Shilon, et al., Application of deep learning methods to analysis of imaging atmospheric Cherenkov telescopes data, Astroparticle Physics 105 (2019) 44-53 [astro-ph/1803.10698]

    work page internal anchor Pith review Pith/arXiv arXiv 2019

  4. [4]

    Chadwick, Very high energy cosmic gamma rays from radio and x-ray pulsars, Ph.D

    P. Chadwick, Very high energy cosmic gamma rays from radio and x-ray pulsars, Ph.D. thesis, University of Durham (1987)

    work page 1987

  5. [5]

    Characterisation and Testing of CHEC-M - a camera prototype for the Small-Sized Telescopes of the Cherenkov Telescope Array

    J. Zorn et al., Characterisation and testing of CHEC-M – A camera prototype for the small-sized telescopes of the Cherenkov telescope array, Nucl. Instrum. Meth. A904 (2018) 44-63 [astro-ph/1806.11308]

    work page internal anchor Pith review Pith/arXiv arXiv 2018

  6. [6]

    Gamma-Hadron Separation in Very-High-Energy gamma-ray astronomy using a multivariate analysis method

    S. Ohm et al., Gamma-Hadron Separation in Very-High-Energyγ-ray astronomy using a multivariate analysis method, Astroparticle Physics 31 (2009) 383-391 [astro-ph/0904.1136]

    work page internal anchor Pith review Pith/arXiv arXiv 2009

  7. [7]

    A Deep Learning-based Reconstruction of Cosmic Ray-induced Air Showers

    M. Erdmann et al., A deep learning-based reconstruction of cosmic ray-induced air showers, Astroparticle Physics 97 (2018) 46–53 [astro-ph/1708.00647]

    work page internal anchor Pith review Pith/arXiv arXiv 2018

  8. [8]

    Nature of the low-energy, {\gamma}-like background for the Cherenkov Telescope Array

    J. Sitarek et al., Nature of the low-energy, γ-like background for the Cherenkov Telescope Array, Astroparticle Physics 97 (2018) 46–53 [astro-ph/1710.06980]

    work page internal anchor Pith review Pith/arXiv arXiv 2018

  9. [9]

    Sahakian,et-al

    V . Sahakian,et-al. Cherenkov light in electron-induced air showers , Astroparticle Physics 25 (2006) 233–241

    work page 2006

  10. [10]

    M. Huennefeld, Deep Learning in Physics exemplified by the Reconstruction of Muon-Neutrino Events in IceCube, in the Proceedings of the 35th International Cosmic Ray Conference, PoS(ICRC2017)1057 (2018)

    work page 2018

  11. [11]

    Chollet, et al., Keras, https://keras.io (2015)

    F. Chollet, et al., Keras, https://keras.io (2015)

    work page 2015

  12. [12]

    CTA Consortium, CTApipe, https://github.com/cta-observatory/ctapipe (2019)

    work page 2019

  13. [13]

    CTA Consortium, DL1-Data-Handler, https://github.com/cta-observatory/dl1-data-handler (2019)

    work page 2019

  14. [14]

    F.Pedregosa et.al., Scikit-learn: Machine Learning in Python, Journal of Machine Learning Research 12 (2011) 2825–2830

    work page 2011

  15. [15]

    Shi et al., Convolutional LSTM network: A machine learning approach for precipitation nowcasting, in the proceedings of Advances in Neural Information Processing Systems 28 (NIPS

    X. Shi et al., Convolutional LSTM network: A machine learning approach for precipitation nowcasting, in the proceedings of Advances in Neural Information Processing Systems 28 (NIPS

    work page

  16. [16]

    Simulation of Imaging Atmospheric Cherenkov Telescopes with CORSIKA and sim_telarray

    K. Bernlöhr, Simulation of imaging atmospheric cherenkov telescopes with corsika and sim_telarray, Astroparticle Physics 30 (2008) 149 – 158 [astro-ph/0808.2253]

    work page internal anchor Pith review Pith/arXiv arXiv 2008

  17. [17]

    O. Le Blanc et al., Final characterisation and design of the Gamma-ray Cherenkov Telescope (GCT) for the Cherenkov Telescope Array, Ground-based and Airborne Telescopes VII 1070010 (2018) 7

    work page 2018