Jet Single Shot Detection

Pol, Adrian Alan; Aarrestad, Thea Klaeboe; Govorkova, Katya; Halily, Roi; Klempner, Anat; Kopetz, Tal; Lončar, Vladimir; Ngadiuba, J.; Pierini, M.; Olya, Sirkin,; Summers, S.

doi:10.1051/epjconf/202125104027

“…The PFJet-SSD architecture is shown in Figure 1. We modify the original SSD architecture [57] and Jet-SSD architecture proposed in [87]. Having in mind an HLT application with a typical latency of ≈ 150 ms, we extend the event image representation to include the information from the charged-particle reconstruction.…”

Section: Methodsmentioning

confidence: 99%

“…Having in mind an HLT application with a typical latency of ≈ 150 ms, we extend the event image representation to include the information from the charged-particle reconstruction. We do so by adding a tracker channel to the image, in front of the calorimeter channels already introduced in [87]. We use a lightweight MobileNet architecture [75] as a backbone for our detector which replaces the convolution operation with a combination of depthwise and pointwise versions.…”

Section: Methodsmentioning

confidence: 99%

“…This is due to the size of the jets. As done in [87] we remove these layers already at the training time. Retaining the deeper layers of the backbone, i.e.…”

Section: Methodsmentioning

confidence: 99%

“…• We generate and publicly share a dataset of simulated LHC collisions, pre-processed to be suited for computer vision applications similar to those discussed in this work, as well as for point-cloud end-to-end reconstruction. The dataset is available on Zenodo [15] and it is accompanied by annotated jet labels, to be used as ground truth during training.…”

Section: Introductionmentioning

confidence: 99%

See 2 more Smart Citations

Lightweight Jet Reconstruction and Identification as an Object Detection Task

Pol

¹

,

Aarrestad

²

,

Говоркова

³

et al. 2022

Preprint

Self Cite

0

View full text Add to dashboard Cite

We apply object detection techniques based on deep convolutional blocks to end-to-end jet identification and reconstruction tasks encountered at the CERN Large Hadron Collider (LHC). Collision events produced at the LHC and represented as an image composed of calorimeter and tracker cells are given as an input to a Single Shot Detection network. The algorithm, named PFJet-SSD performs simultaneous localization, classification and regression tasks to cluster jets and reconstruct their features. This all-in-one single feed-forward pass gives advantages in terms of execution time and an improved accuracy w.r.t. traditional rule-based methods. A further gain is obtained from network slimming, homogeneous quantization, and optimized runtime for meeting memory and latency constraints of a typical real-time processing environment. We experiment with 8-bit and ternary quantization, benchmarking their accuracy and inference latency against a single-precision floating-point. We show that the ternary network closely matches the performance of its full-precision equivalent and outperforms the state-of-the-art rule-based algorithm. Finally, we report the inference latency on different hardware platforms and discuss future applications.

show abstract

“…• We generate and publicly share a dataset of simulated LHC collisions, pre-processed to be suited for computer vision applications similar to those discussed in this work, as well as for point-cloud end-to-end reconstruction. The dataset is available on Zenodo [16] and it is accompanied by annotated jet labels, to be used as ground truth during training.…”

Section: Introductionmentioning

confidence: 99%

Lightweight jet reconstruction and identification as an object detection task

Pol¹,

Aarrestad²,

Govorkova³

et al. 2022

Mach. Learn.: Sci. Technol.

Self Cite

2

0

View full text Add to dashboard Cite

We apply object detection techniques based on deep convolutional blocks to end-to-end jet identification and reconstruction tasks encountered at the CERN Large Hadron Collider (LHC). Collision events produced at the LHC and represented as an image composed of calorimeter and tracker cells are given as an input to a Single Shot Detection network. The algorithm, named PFJet-SSD performs simultaneous localization, classification and regression tasks to cluster jets and reconstruct their features. This all-in-one single feed-forward pass gives advantages in terms of execution time and an improved accuracy w.r.t. traditional rule-based methods. A further gain is obtained from network slimming, homogeneous quantization, and optimized runtime for meeting memory and latency constraints of a typical real-time processing environment. We experiment with 8-bit and ternary quantization, benchmarking their accuracy and inference latency against a single-precision floating-point. We show that the ternary network closely matches the performance of its full-precision equivalent and outperforms the state-of-the-art rule-based algorithm. Finally, we report the inference latency on different hardware platforms and discuss future applications.

show abstract

Sharpening the A → Z(*)h signature of the Type-II 2HDM at the LHC through advanced Machine Learning

Esmail,

Hammad,

Moretti

2023

J. High Energ. Phys.

3

0

View full text Add to dashboard Cite

The A → Z(*)h decay signature has been highlighted as possibly being the first testable probe of the Standard Model (SM) Higgs boson discovered in 2012 (h) interacting with Higgs companion states, such as those existing in a 2-Higgs Doublet Model (2HDM), chiefly, a CP-odd one (A). The production mechanism of the latter at the Large Hadron Collider (LHC) takes place via $$ b\overline{b} $$ b b ¯ -annihilation and/or gg-fusion, depending on the 2HDM parameters, in turn dictated by the Yukawa structure of this Beyond the SM (BSM) scenario. Among the possible incarnations of the 2HDM, we test here the so-called Type-II, for a twofold reason. On the one hand, it intriguingly offers two very distinct parameter regions compliant with the SM-like Higgs measurements, i.e., where the so-called ‘SM limit’ of the 2HDM can be achieved. On the other hand, in both configurations, the AZh coupling is generally small, hence the signal is strongly polluted by backgrounds, so that the exploitation of Machine Learning (ML) techniques becomes extremely useful. In this paper, we show that the application of advanced ML implementations can be decisive in establishing such a signal. This is true for all distinctive kinematical configurations involving the A → Z(*)h decay, i.e., below threshold (mA < mZ + mh), at its maximum (mZ + mh < mA < 2mt) and near the onset of $$ t\overline{t} $$ t t ¯ pair production (mA ≈ 2mt), for which we propose Benchmark Points (BPs) for future phenomenological analyses.

show abstract

Jet Single Shot Detection

Cited by 4 publications

References 52 publications

Lightweight Jet Reconstruction and Identification as an Object Detection Task

Lightweight Jet Reconstruction and Identification as an Object Detection Task

Lightweight jet reconstruction and identification as an object detection task

Sharpening the A → Z(*)h signature of the Type-II 2HDM at the LHC through advanced Machine Learning

Contact Info

Product

Resources

About