Pith. sign in

REVIEW 2 major objections 6 minor 31 references

Compact decision-tree classifiers in the Belle II drift chamber front-end FPGA cut cross-talk noise by about half while keeping signal efficiency above 98 percent, and in calibration runs reduced track-trigger rates by up to around 50 perce

Reviewed by Pith at T0; open to challenge. T0 means a machine referee read the full paper against a public rubric. the ladder, T0–T4 →

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗

T0 review · deepseek-v4-flash

2026-08-01 03:11 UTC pith:7TWN43RM

load-bearing objection First real ML-in-FEE deployment for Belle II CDC noise suppression; online numbers are convincing, but the offline factor-of-two claim needs a documented train/test split. the 2 major comments →

arxiv 2607.25174 v1 pith:7TWN43RM submitted 2026-07-28 physics.ins-det hep-ex

Machine-Learning-Based Waveform Discrimination in the Front-End Electronics of the Belle II Central Drift Chamber for Cross-Talk Noise Reduction

Yun-Tsung Lai , Taichiro Koga , Yu Nakazawa , Nanae Taniguchi , Keisuke Yoshihara This is my paper
classification physics.ins-det hep-ex
keywords boosted decision treescross-talk noiseFPGA inferencefront-end electronicswaveform discriminationBelle II central drift chamberLevel-1 trigger
verification ladder T0 review T1 audit T2 compute T3 formal T4 reserved

The pith

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

This paper tries to prove that cross-talk noise in the Belle II central drift chamber can be stopped at its source, inside the front-end electronics, by classifying each wire's ADC waveform with a tiny machine-learning model. The classifiers are boosted decision trees small enough to fit into the Virtex-5 FPGA on each front-end board, where they run in a fully pipelined way with a latency of two clock cycles. Offline, the models reject roughly half of cross-talk hits while keeping signal efficiency above 98 percent. In dedicated calibration runs with the actual firmware, track-segment and track-trigger rates fell by up to about 50 percent, and trigger acceptance for events containing real tracks stayed within 10 percent of the baseline. If this holds at full luminosity, it provides a template for moving intelligent waveform processing into the front-end electronics of future detectors.

Core claim

The central claim is that the full ADC waveform, available only in the drift chamber's front-end electronics, contains enough information to distinguish real charged-track hits from FEE cross-talk noise, and that a suitably compact boosted-decision-tree classifier can exploit that information in real time on a limited FPGA. The paper reports that five waveform samples centered on the TDC timing, quantized to eight bits, feed four depth-four decision trees per wire channel; 96 such models cover the 48 channels in the two chamber geometries. Offline this configuration cuts cross-talk noise by roughly a factor of two at a signal efficiency above 98 percent. When deployed in firmware, it reduced

What carries the argument

The central mechanism is a per-channel Boosted Decision Tree (BDT) classifier, an ensemble of four shallow decision trees whose outputs are summed, converted to a custom two-stage pipelined VHDL implementation in the FEE's Virtex-5 FPGA. All decision-node conditions are evaluated in parallel in one clock cycle and leaf outputs combined with AND gates in the next, giving 15.7 ns inference. Five ADC samples around each TDC hit, truncated to eight bits, are the input features; independent models are trained per wire channel and for the two chamber geometries (SL0 vs SL1–SL8). The design keeps TDC timing flowing to the trigger immediately and attaches the BDT classification later at the Track Se

Load-bearing premise

Training and validation waveforms came from a 2024 calibration run and a March 2026 run with only an electron beam at 1 A; the paper assumes these faithfully represent cross-talk noise under future high-luminosity beam-beam collisions, and it states the cross-talk origin is not yet fully understood.

What would settle it

Record full ADC waveforms during regular beam-beam operation at design luminosity and measure the deployed BDT's cross-talk rejection rate and 2D Finder trigger rate; if the factor-of-two rejection and roughly 50 percent trigger-rate reduction do not reproduce, or if track acceptance drops by more than about 10 percent, the central claim would be weakened.

Watch this falsifier — get emailed when new claim-graph text bears on it.

If this is right

  • Cross-talk wire hits can be cut by roughly half at the front end while retaining more than 98 percent of genuine signal hits.
  • Reduced wire-hit rates translate into track-segment and Level-1 trigger rate reductions of up to about 50 percent in calibration runs, with the 2D Finder rate falling from 82.6 kHz to 23.3 kHz at the stricter cut.
  • Track trigger acceptance for events with reconstructed IP tracks is preserved, with losses below about 10 percent.
  • The added power consumption on the front-end board is about 2 percent, and inference latency is two clock cycles (15.7 ns), so the approach fits the existing 4.4 microsecond trigger budget.
  • Channel-wise independent waveform classification is sufficient, so no inter-channel data sharing is needed in the FPGA.

Where Pith is reading between the lines

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

  • Editorial inference: The training and validation waveforms came from a 2024 calibration run and a March 2026 run with only an electron beam at 1 A; the paper does not measure performance under full beam-beam background, so the deployed thresholds may need retraining if cross-talk shapes change with luminosity.
  • Editorial inference: Because the BDT uses only five samples per hit, its discrimination is effectively a template on pulse shape and timing; a natural extension the paper itself identifies is using neighboring-channel information via CNN or GNN architectures.
  • Editorial inference: The approach could generalize beyond cross-talk: any front-end electronics with access to raw waveforms could use compact BDTs for pile-up separation, baseline correction, or other preprocessing, provided the FPGA resources are comparable.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit.

Referee Report

2 major / 6 minor

Summary. This paper reports the development and deployment of compact BDT-based waveform classifiers in the front-end FPGAs of the Belle II CDC to suppress cross-talk noise before it reaches the Level-1 trigger. The authors train 96 channel-specific quantized BDTs using five ADC samples around each TDC hit, implement a two-cycle custom VHDL inference engine using ~4.5% of the Virtex-5 resources, and validate it in dedicated March 2026 single-beam calibration runs. Offline processing (Table II) reports signal efficiencies above ~98% with roughly half of cross-talk hits rejected; the online runs show 48–72% reductions in the 2D Finder trigger rate and ~50% reductions in outer-super-layer track-segment rates, while event-level acceptance for events containing IP tracks is preserved at roughly the 90% level.

Significance. If the offline wire-level numbers survive a proper holdout evaluation, this is a valuable, experimentally demonstrated use of ML in detector front-end electronics under tight resource and latency constraints. The main strengths are the real FPGA deployment in an operating experiment, the measured online trigger-rate reduction, the explicit statement of resource and power usage, the channel-by-channel quantization study, and the very small degradation from quantization. The main concern is that the wire-level offline table is the quantitative basis for the headline 'factor of two / >98%' claim, but its train/test separation is not documented. The acceptance statement for one trigger also needs a small correction relative to Table V.

major comments (2)
  1. [§IV-A, Table II] The offline wire-level performance is the basis for the factor-of-two cross-talk rejection and the >98% signal efficiency, but the text does not state that the evaluation set excludes the samples used to train the 96 BDT models. Section IV-A says 'the performance was evaluated using all waveform samples' and the cut thresholds are 'determined from their ROC curves' without indicating whether those curves come from the same data. Given ~10k training samples per class and 96 channel-specific models, in-sample evaluation could inflate the reported rejection rates. Please add an explicit train/test split (e.g., a held-out run, time-ordered split, per-channel holdout, or k-fold CV) and report Table II on the held-out data, including the number of samples in the evaluation set.
  2. [Abstract/Conclusion vs. Table V] The claim that track-trigger acceptance for events with tracks is preserved 'within 10%' is not supported for both triggers at cut level 2. Table V shows the Neural 3D Tracker acceptance for events with IP tracks changing from 0.790±0.032 (no cut) to 0.702±0.027 (cut 2), a relative loss of about 11.1%; only the 2D Finder (5.1% relative loss) satisfies the strict 10% statement. Please qualify the claim (e.g., 'approximately 10%'), combine both triggers into a single metric, or discuss the uncertainty and why the 11% loss is acceptable.
minor comments (6)
  1. [§II-A/II-B/II-C] The paper states 14,336 anode wires and 48 channels per FEE board, then mentions 292 FEE boards. 292×48 = 14,016, leaving 320 wires unaccounted for; please check the number of boards or the per-board channel count.
  2. [§III-B/IV-A] The 'consecutive increments' defining cut levels 2–4 are not specified. Please give the actual threshold values or the exact rule used to move from one cut level to the next, so the deployed cuts are reproducible.
  3. [§IV-A] For waveform samples with multiple TDC hits, the paper states that a wire passes if at least one corresponding classifier output exceeds the threshold. This OR rule makes the selection more permissive and should be justified; it also affects how the reported signal and background efficiencies should be interpreted.
  4. [Table II] No uncertainties are reported for the efficiencies/rejection rates in Table II. With 96 models and finite samples, per-channel spread or binomial confidence intervals would help assess whether the differences between cut levels and between SL0 and SL1–SL8 are meaningful.
  5. [§III-C/Table I] The Conifer HLS comparison is obtained on a Kintex-7 device because HLS does not support Virtex-5. The caption notes this, but the text should explicitly warn that the resource numbers for Conifer HLS are not directly comparable to the Virtex-5 implementation.
  6. [Conclusion] The sentence 'a loss of less than 10% loss' contains a typo; should be 'a loss of less than 10%'.

Circularity Check

1 steps flagged

No derivation-level circularity; offline wire-level numbers are likely in-sample, but the March 2026 calibration-run validation independently confirms rate reductions.

specific steps
  1. fitted input called prediction [Section IV-A 'Wire-Level Performance Based on Offline Waveform Data', Table II]
    "Although only single-TDC samples were used during training, the performance was evaluated using "all waveform samples." ... To define common selection criteria for all 96 quantized BDT models, the classifier thresholds were determined from their ROC curves. The threshold corresponding to a signal efficiency of approximately 99% was defined as "cut level 1.""

    The paper evaluates on all waveform samples without explicitly excluding training samples, and sets cut 1 from the same ROC curves, so the reported ~99% signal efficiency at cut 1 is essentially the operator-selected operating point, not an independent prediction. The rejection rates in Table II are resubstitution estimates. The online calibration runs use new data and independently confirm rate reductions, keeping overall circularity low.

full rationale

The paper is empirical: BDT classifiers are trained on recorded waveforms and validated in March 2026 calibration runs with new data. No claim reduces to a fitted parameter renamed as prediction, no load-bearing self-citation, no import of a uniqueness theorem. The 2D Finder rate fell 82.6 to 23.3 kHz at cut 2 with track acceptance within 10%, which is independent evidence. The only hazard is the offline Table II evaluation lacks an explicit train/test split and thresholds are chosen from the same ROC curves, so the >98% efficiency and rejection rates may be optimistic. This minor circularity in reporting does not invalidate the online validation.

Axiom & Free-Parameter Ledger

3 free parameters · 4 axioms · 0 invented entities

The paper introduces no new physics entities. The central claims depend on operational free parameters (thresholds, hyperparameters, feature window) fitted to data, and on domain assumptions about label purity and the representativeness of the calibration conditions.

free parameters (3)
  • BDT selection thresholds (cut levels 1-4) = Cut level 1 set to ~99% signal efficiency on training ROC; cut levels 2-4 by successive threshold increments
    Chosen from the ROC curves of the quantized models on waveform data; they directly determine the reported rejection/efficiency trade-off. They are fitted to the training data, not derived.
  • BDT hyperparameters (n_estimators, max_depth, learning_rate) = 4 trees, max depth 4, learning rate 1.0
    Hand-selected to meet FPGA resource/latency constraints; no systematic optimization is reported.
  • Input feature window (5 ADC samples centered on TDC timing) = 5 samples at 31.4 MSPS
    Manually chosen feature representation; other window sizes/alignments were not reported.
axioms (4)
  • domain assumption Cross-talk noise can be distinguished from real track signals based on a single wire's 5-point ADC waveform.
    The whole method depends on waveform shape carrying discriminative information for this specific noise; the paper notes the cross-talk origin is not fully understood (Sec II-D).
  • domain assumption Track-associated hits with fewer than 4 fired wires in an ASIC are pure signal, and non-associated hits with more than 6 fired wires are pure cross-talk.
    Used to define training labels (Sec III-B); if the 4-6 wire middle region or misassociation contaminates the labels, the classifier is trained on noisy labels.
  • domain assumption The March 2026 single-electron-beam calibration runs are representative of noise conditions in full-luminosity collisions.
    The validation was only performed with a 1 A electron beam, not with beam-beam collisions at design luminosity (Sec IV-B); extrapolation relies on this assumption.
  • standard math Standard ML/BDT background results (scikit-learn GradientBoostingClassifier, quantization) work as expected.
    The paper relies on standard libraries; no formal proof needed.

pith-pipeline@v1.3.0-alltime-deepseek · 10707 in / 11195 out tokens · 100518 ms · 2026-08-01T03:11:08.141216+00:00 · methodology

0 comments
read the original abstract

Machine learning (ML) inference on FPGAs has been widely adopted in real-time triggering of collider experiments for detector signature identification. In contrast, the ML application in Front-End Electronics (FEE) has not yet been fully explored, primarily due to constraints such as limited FPGA resources, power consumption, and localized detector coverage. In this work, we develop an ML-based waveform discrimination method for the Central Drift Chamber (CDC) of the Belle II experiment to suppress cross-talk noise at the front-end level. The Belle II CDC is a key charged-particle tracking detector for both offline and the real-time hardware trigger. During Belle II operation, background wire hits have been observed in the CDC FEE, where multiple hits occur in neighboring anode wires by large energy deposit. The hardware track trigger employs a Hough transformation based on track segments formed by combining hits from multiple wire layers. Due to the reduced information, the track trigger is sensitive to cross-talk noise, hence resulting in an increased fake trigger rate with higher luminosity in the future. We employ compact and fast Boosted Decision Tree models implemented in a Xilinx Virtex-5 FPGA of the CDC FEE, where waveform is processed independently for each wire channel in a fully pipelined manner. Offline studies show that the cross-talk noise can be reduced by approximately a factor of two while maintaining a signal efficiency above 98%. The firmware validation during dedicated Belle II calibration runs demonstrated reductions of up to 50% in track segment and trigger rates while preserving the trigger acceptance for events containing tracks within 10%. This work demonstrates the technical feasibility of compact and low-latency ML inference in detector FEE and highlights its potential for future intelligent detector readout systems in high-energy physics experiments.

Figures

Figures reproduced from arXiv: 2607.25174 by Keisuke Yoshihara, Nanae Taniguchi, Taichiro Koga, Yu Nakazawa, Yun-Tsung Lai.

Figure 1
Figure 1. Figure 1: The mechanical design of the Belle II CDC detector. [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: Wire configuration of the Belle II CDC detector. [PITH_FULL_IMAGE:figures/full_fig_p002_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: Front-End Electronics device of Belle II CDC. [PITH_FULL_IMAGE:figures/full_fig_p003_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: The shapes of the track segments for tracking with CDC in Level-1 [PITH_FULL_IMAGE:figures/full_fig_p003_4.png] view at source ↗
Figure 6
Figure 6. Figure 6: Number of fired wires in the corresponding ASIC. [PITH_FULL_IMAGE:figures/full_fig_p004_6.png] view at source ↗
Figure 7
Figure 7. Figure 7: Examples of ADC waveform of fired wire channels in CDC. The [PITH_FULL_IMAGE:figures/full_fig_p004_7.png] view at source ↗
Figure 8
Figure 8. Figure 8: The receiver operating characteristic curve of one of the 96 quantized [PITH_FULL_IMAGE:figures/full_fig_p005_8.png] view at source ↗
Figure 9
Figure 9. Figure 9: Track segment rates as a function of the azimuthal angle ( [PITH_FULL_IMAGE:figures/full_fig_p006_9.png] view at source ↗

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

31 extracted references · 4 canonical work pages

  1. [1]

    The Belle detector,

    A. Abashianet al.(Belle Collaboration), “The Belle detector,”Nucl. Instrum. Methods Phys. Res., Sect. A,479, 117, Feb. 2002

    2002

  2. [2]

    Overview of the KEKB accelerators,

    S. Kurokawa and E. Kikutani, “Overview of the KEKB accelerators,” Nucl. Instrum. Methods Phys. Res., Sect. A,499, 1, Feb. 2003. IEEE TRANSACTIONS ON NUCLEAR SCIENCE, VOL. XX, NO. XX, XXXX 2026 8

    2003

  3. [3]

    Achievements of KEKB,

    T.Abeet al., “Achievements of KEKB,”Prog. Theor. Exp. Phys.,2013, 03A001, Mar. 2013

    2013

  4. [4]

    Belle II technical design report,

    T. Abeet al., “Belle II technical design report,” 2010,arXiv:1011.0352

    Pith/arXiv arXiv 2010

  5. [5]

    SuperKEKB collider,

    K. Akai, K. Furukawa, and H. Koiso, “SuperKEKB collider,”Nucl. Instrum. Methods Phys. Res. A, Accel. Spectrom. Detect. Assoc. Equip., vol. 907, pp. 188–199, Nov. 2018

    2018

  6. [6]

    SuperB Collaboration, SuperB: A High-Luminosity Asymmetric e+ e- Super Flavor Factory, Conceptual Design Re- port,

    M. Bonaet al., “SuperB Collaboration, SuperB: A High-Luminosity Asymmetric e+ e- Super Flavor Factory, Conceptual Design Re- port,” SLAC-R-856,arxiv:0709.0451, [physics.hep-ex] INFN-AE-07-02, LAL-07-15, INFN-AE-07-2, Sep. 2007

    Pith/arXiv arXiv 2007

  7. [7]

    The Belle II Physics Book,

    E. Kou,et al., “The Belle II Physics Book,”arXiv:1808.10567, [physics.hep-ex]Prog Theor Exp Phys, Dec. 2019

    arXiv 2019

  8. [8]

    Level 1 trigger system for the Belle II experiment,

    Y . Iwasaki, B. Cheon, E. Won, and G. Varner, “Level 1 trigger system for the Belle II experiment,”IEEE Trans. Nucl. Sci., vol. 58, no. 4, pp. 1807–1815, Aug. 2011

    2011

  9. [9]

    Data acquisition system for the Belle II experiment,

    S. Yamadaet al., “Data acquisition system for the Belle II experiment,” IEEE Trans. Nucl. Sci., vol. 62, no. 3, pp. 1175–1180, Jun. 2015

    2015

  10. [10]

    Data acquisition system for Belle II,

    M. Nakao, T. Higuchi, R. Itoh, and S. Y . Suzuki, “Data acquisition system for Belle II,”J. Instrum., vol. 5, no. 12, Dec. 2010, Art. no. C12004, doi: 10.1088/1748-0221/5/12/c12004

    work page doi:10.1088/1748-0221/5/12/c12004 2010

  11. [11]

    Central drift chamber for Belle-II,

    N. Taniguchi, “Central drift chamber for Belle-II,”J. Instrum., vol. 12, no. 6, Jun. 2017., Art. no. C06014, doi: 10.1088/1748- 0221/12/06/C06014

    doi:10.1088/1748- 2017

  12. [12]

    Inner chamber of the Belle II Central Drift Chamber,

    K. Chaiwongkhotet al., “Inner chamber of the Belle II Central Drift Chamber,”2013 IEEE Nuclear Science Symposium and Medical Imaging Conference (2013 NSS/MIC), Seoul, Korea (South), pp. 1–3, 2013, doi: 10.1109/NSSMIC.2013.6829461

    arXiv 2013

  13. [13]

    Front-end electronics of the Belle II drift chamber,

    S. Shimazakiet al., “Front-end electronics of the Belle II drift chamber,” Nucl. Instrum. Methods Phys. Res., Sect. A,735, 193, Jan. 2014

    2014

  14. [14]

    Virtex-5 Family Overview,

    AMD Xilinx, “Virtex-5 Family Overview,” https://docs.amd.com/v/u/en- US/ds100

  15. [15]

    Virtex-5 FPGA RocketIO GTP Transceiver,

    AMD Xilinx, “ Virtex-5 FPGA RocketIO GTP Transceiver,” https://docs.amd.com/v/u/en-US/ug196

  16. [16]

    Virtex-6 Family Overview,

    AMD Xilinx, “Virtex-6 Family Overview,” https://docs.amd.com/v/u/en- US/ds150

  17. [17]

    UltraScale Architecture and Product Data Sheet: Overview,

    AMD Xilinx, “UltraScale Architecture and Product Data Sheet: Overview,” https://docs.amd.com/v/u/en-US/ds890-ultrascale-overview

  18. [18]

    The level 1 trigger system for Belle II CDC,

    J.-G. Shiu, “The level 1 trigger system for Belle II CDC,”2016 IEEE- NPSS Real Time Conference, RT, Padua, Italy, pp. 1–3, 2016, doi: 10.1109/RTC.2016.7543137

    arXiv 2016

  19. [19]

    Development of the level-1 track trigger with central drift chamber detector in Belle II experiment and its performance in SuperKEKB 2019 phase 3 operation,

    Y .-T. Laiet al., “Development of the level-1 track trigger with central drift chamber detector in Belle II experiment and its performance in SuperKEKB 2019 phase 3 operation,”J. Instrum., vol. 15, no. 6, Jun. 2020, Art. no. C06063, doi: 10.1088/1748-0221/15/06/C06063

    work page doi:10.1088/1748-0221/15/06/c06063 2019

  20. [20]

    An overview of the hardware trigger system for the drift chamber of Belle II,

    H. Baeet al., “An overview of the hardware trigger system for the drift chamber of Belle II,”J. Instrum., vol. 21, no. 5, Nov. 2025, Art. no. C05016, doi: 10.1088/1748-0221/21/05/C05016

    work page doi:10.1088/1748-0221/21/05/c05016 2025

  21. [21]

    The Neural Network First-Level Hardware Track Trigger of the Belle II Experiment,

    S. B ¨ahret al., “The Neural Network First-Level Hardware Track Trigger of the Belle II Experiment,” 2024,arXiv:2402.14962[hep-ex]

    Pith/arXiv arXiv 2024

  22. [22]

    Design of the Global Reconstruction Logic in the Belle II Level-1 Trigger system,

    Y .-T. Laiet al., “Design of the Global Reconstruction Logic in the Belle II Level-1 Trigger system,”Nucl. Instrum. Methods Phys. Res., Sect. A, 1078, 170577, Sep. 2025

    2025

  23. [23]

    Upgrade of Two Dimensional Track Trigger on Central Drift Chamber aimed for Belle II Targeted Luminosity,

    P. Ni, “Upgrade of Two Dimensional Track Trigger on Central Drift Chamber aimed for Belle II Targeted Luminosity,” M.S. thesis, Depart- ment of Physics, Graduate School of Science, The University of Tokyo, Tokyo, Japan, Aug. 2022

    2022

  24. [24]

    Hardware-accelerated GNN-based hit filtering for the Belle II Level-1 trigger,

    G. Heineet al., “Hardware-accelerated GNN-based hit filtering for the Belle II Level-1 trigger,”J. Instrum., vol. 21, no. 2, Feb. 2026, Art. no. C02007, doi: 10.1088/1748-0221/21/02/C02007

    work page doi:10.1088/1748-0221/21/02/c02007 2026

  25. [25]

    Belle II Tracking Group Collaboration, Track finding at Belle II,

    V . Bertacchiet al., “Belle II Tracking Group Collaboration, Track finding at Belle II,”Comput. Phys. Comm., vol. 259. pp. 107610, Feb. 2021

    2021

  26. [26]

    Belle II software,

    T. Kuhret al., “Belle II software,”J. Phys.: Conf. Ser., vol. 762, no. 012007,

  27. [27]

    The Belle II core software,

    T. Kuhret al., “The Belle II core software,”,Comput. Softw. Big Sci., vol. 3, no. 1, Nov. 2018

    2018

  28. [28]

    Fast inference of Boosted Decision Trees in FPGAs for particle physics,

    S. Summerset al., “Fast inference of Boosted Decision Trees in FPGAs for particle physics,”J. Instrum., vol. 15, no. 5, May 2020, Art. no. P05026, doi: 10.1088/1748-0221/15/05/P05026

    doi:10.1088/1748-0221/15/05/p05026 2020

  29. [29]

    The slow control and data quality monitoring system for the Belle II experiment

    T. Konnoet al., “The slow control and data quality monitoring system for the Belle II experiment”,IEEE Trans. Nucl. Sci., vol. 62, no. 3, pp. 897–902, 2015

    2015

  30. [30]

    Trigger slow control system of the Belle II experi- ment,

    C.-H. Kimet al., “Trigger slow control system of the Belle II experi- ment,”Nucl. Instrum. Methods Phys. Res., Sect. A,1014, 165748, Oct. 2021

    2021

  31. [31]

    Machine Learning on Heterogeneous, Edge, and Quantum Hardware for Particle Physics (ML-HEQUPP),

    J. Gonskiet al., “Machine Learning on Heterogeneous, Edge, and Quantum Hardware for Particle Physics (ML-HEQUPP),” 2026, arXiv:2602.22248[physics.ins-det]

    arXiv 2026