Particle transformers for identifying Lorentz-boosted Higgs bosons decaying to a pair of W bosons

summary

The gist

Given the high stakes involved, I will ensure this synthesis is precise and thorough.

In short

A novel deep neural network called Particle Transformer (PaRT) was developed to identify highly boosted resonances like Higgs bosons decaying into W bosons, which appear as single, complex jets. PaRT uses a self-attention mechanism to weigh particle importance. It successfully tagged these jets with over 50% efficiency against only 1% background noise, significantly boosting sensitivity for new physics searches.

Key concepts

Particle Transformer (PaRT)
PaRT is a deep neural network classifier designed specifically for identifying highly boosted resonances reconstructed as single, multipronged jets. It uses a self-attention mechanism to dynamically assess the significance of different particles within the jet to improve classification accuracy.
Self-Attention Mechanism
This is an architectural feature in the neural network that allows PaRT to weigh the importance of every particle within a jet simultaneously. It enables the model to focus on the most relevant constituents needed to distinguish a specific resonance from background noise, which is crucial for complex boosted systems.
Tagging Efficiency and Background Efficiency
These metrics measure how well PaRT can correctly identify a target signal (tagging efficiency) versus how often it mistakenly identifies random noise as a signal (background efficiency). The paper reports over 50% tagging efficiency while keeping background contamination very low at only 1%, indicating high purity.
Lorentz-Boosted Resonances
These are particles or systems moving at extremely high speeds, causing them to appear highly compressed and merged when they interact with detectors. The PaRT algorithm is specifically designed to handle these 'boosted' signals, which are characteristic of heavy particles like the Higgs boson decaying into W bosons.

Terminology used across episodes

This episode discusses

The paper

Particle transformers for identifying Lorentz-boosted Higgs bosons decaying to a pair of W bosons · Read on arXiv

The CMS Collaboration

European Organization for Nuclear Research (CERN)

A novel deep neural network classifier, a ``Particle transformer'' (PaRT), is introduced for the identification of highly Lorentz-boosted resonances reconstructed as single, multipronged jets in measurements and searches performed by the CMS Collaboration at the CERN LHC. Based on a self-attention mechanism that allows the model to weigh the importance of different particles, PaRT is trained on a wide variety of topologies, notably demonstrating strong performance for the first time on jets originating from boosted Higgs boson decays to W bosons. The PaRT algorithm achieves a tagging efficiency of more than 50% for such jets at a background efficiency of 1%, while maintaining decorrelation from the jet mass. A calibration is performed in proton-proton collision data collected by CMS at a center-of-mass energy of 13 TeV, with a data set corresponding to a total luminosity of 138 fb-1. Data-to-simulation selection efficiency scale factors are measured to be in the 0.9 - 1.0 range, with relative uncertainties between 7 and 23%. The tagging capability of PaRT enhances the sensitivity of standard model measurements and searches for beyond-the-standard-model resonances decaying to hadronic diboson systems.

DOI: 10.1007/JHEP10(2026)167

Transcript

Introduction to the show: ident: High Energy Physics: Experiment. The day in high-energy experiment: detectors, searches and limits.

Kai: We're looking at the paper: "Particle transformers for identifying Lorentz-boosted Higgs bosons decaying to a pair of W bosons".

Mira: Here's the gist of the paper. Given the high stakes involved, I will ensure this synthesis is precise and thorough.

Kai: Detailed Research Summary: Particle Transformer (PART) Algorithm The Particle Transformer (PART) is a novel deep neural network classifier introduced for the identification of highly Lorentz-boosted resonances reconstructed as single, multipronged jets within the measurements and searches conducted by the CMS Collaboration at the CERN Large Hadron Collider (LHC).

Mira: Core Methodology: At its foundation, PART is built upon a sophisticated self-attention mechanism.

Kai: This architecture grants the model the crucial capability to dynamically weigh the importance of different constituent particles within a jet, allowing it to effectively capture complex topological features inherent in boosted systems.

Paper summary: Kai: So, to recap, this paper introduces the Particle Transformer, PART, which uses a self-attention mechanism for classifying highly Lorentz-boosted resonances reconstructed as single jets. The main claim is that it works well on jets from Higgs boson decays into W bosons.

Mira: It's about addressing the challenge of identifying these boosted particles amidst a lot of standard model background, especially because they are so boosted in the high-energy regime.

Lev: I wonder if this self-attention approach adds significant computational overhead compared to, say, a simpler convolutional network structure for jet classification?

Kai: The paper states that PART is trained on a wide variety of topologies to ensure it learns these complex features well. It's not just one kind of jet it's trained on.

Mira: And the performance metric they highlight is that over fifty percent tagging efficiency for those Higgs jets at a background efficiency of one percent, which suggests a very clean separation.

Lev: If we were to put this on real hardware, would the complexity of the self-attention mechanism allow for fast enough inference times to keep up with real-time data streams?

Kai: The paper doesn't explicitly state inference time benchmarks, but they focus heavily on the classification and regression tasks together. They also mention training objectives where the classification uses a cross-entropy loss function, and the jet mass regression uses a log-cosh loss function.

Mira: That dual task—classification plus mass regression—that’s what makes it robust because it forces the model to learn both *what* the particle is and *how much* momentum it has.

Lev: And how does that mass regression part actually translate to real hardware? Does that loss function demand a very precise estimation of the true mass, or is there some tolerance?

Kai: The paper says they use a generator-level mass, called m SD, for QCD jets in the regression task. That's an approximation based on the simulation used during training.

Mira: So we have to trust that the simulation accurately represents how a real jet mass behaves when it's going through this boost effect.

Lev: It sounds like the main assumption here is that our current understanding of jet substructure and its dependence on boost is sufficient for this model to generalize well beyond the training set.

Conclusion: Kai: So, looking at the Particle Transformers for identifying Lorentz-boosted Higgs bosons decaying to a pair of W bosons paper, it’s about building a smarter way to spot those super boosted particles in experiments like CMS at the LHC.

Mira: The authors are aiming to improve how we measure and search for things that go beyond the standard model by getting better at handling these tricky final states.

Lev: It seems like this work moves the needle on experimental sensitivity because it allows us to see signals that were previously buried under the noise of QCD multijet events.

Kai: Exactly, they're saying we can get a cleaner signal for Higgs decays into W bosons by using these new deep learning tools instead of older methods.

Mira: The real significance lies in how this technique helps us probe physics that standard models can't easily explain on their own, like new particles decaying into these hadronic systems.

Lev: So, what we're hearing is that the computational cost might be high, but the potential gain in sensitivity for finding new physics signals is worth investigating further.

Kai: It’s definitely a tool to keep pushing the boundaries of what we can reconstruct and identify at a high-energy collider.

More episodes

← Home