Selected work

Representative Papers

Robust Prediction Variance Estimation for Gaussian Process Regression Under Covariance Smoothness Misspecification

Jun 02, 2026

This study addresses the underestimation of predictive variance in Gaussian process regression caused by misspecification of the covariance function’s smoothness, which induces bias in the mean squared prediction error (MSPE) estimation of the empirical best linear unbiased predictor (EBLUP). The authors establish, for the first time, that when the measures induced by the true and assumed covariance functions are mutually singular, the MSPE bias converges to a strictly positive limit that varies smoothly with the prediction location. Building on this insight, they propose a novel robust MSPE estimator that explicitly accounts for covariance uncertainty. Both theoretical analysis and numerical experiments demonstrate that the proposed estimator substantially outperforms four state-of-the-art alternatives across various smoothness misspecification scenarios, with its advantage becoming more pronounced as the degree of misspecification increases.

0 citationsRead paper

B-jet Tagging Using a Hybrid Edge Convolution and Transformer Architecture

Mar 22, 2026

This work proposes the Edge Convolution Transformer (ECT) model to enhance the discrimination of b-jets from c-jets and light-flavor jets in hadronic collisions. ECT uniquely integrates edge convolution with Transformer self-attention to jointly model track-level features—such as impact parameters and momentum significance—and jet-level observables, including vertex and kinematic variables. This architecture preserves local geometric structure while capturing long-range dependencies. Evaluated on ATLAS simulation data, ECT achieves an AUC of 0.9333, substantially outperforming ParticleNet (0.8904) and a pure Transformer baseline (0.9216), with notable gains in rejecting the most challenging c-jet background. Moreover, its per-jet inference latency remains below 0.060 milliseconds, satisfying the real-time triggering requirements of the LHC.

0 citationsRead paper

MEDIC: a network for monitoring data quality in collider experiments

Nov 22, 2025

In high-energy collider experiments, conventional data quality monitoring (DQM) heavily relies on real detector data, suffering from detector-specific constraints, high data acquisition costs, and extensive manual intervention—making it ill-suited for the scale and complexity of next-generation experiments. This work proposes a simulation-driven machine learning DQM framework: leveraging an enhanced Delphes fast simulator to generate controllable, labeled anomalous events; training neural networks for event-level integrity and consistency monitoring; and enabling subsystem-level fault localization. The approach decouples DQM from real experimental data, ensuring cross-detector transferability and strong scalability. Preliminary validation demonstrates that the model efficiently detects and localizes representative detector faults under simulated anomaly scenarios. This constitutes the first simulation–learning co-design paradigm for generalized, automated DQM in particle physics, paving the way for robust, scalable monitoring in future large-scale experiments.

0 citationsRead paper

Jet Image Tagging Using Deep Learning: An Ensemble Model

Aug 09, 2025

Jet substructure classification in high-energy physics is critical for discovering new physics, yet its complex, high-dimensional nature limits the performance of conventional approaches. This paper proposes a novel jet classification framework based on image-like representation and deep ensemble learning: particle-flow data are encoded as two-dimensional histograms, and a dual-branch neural network ensemble is constructed to jointly support both binary (e.g., top quark vs. light quark) and multi-class (top/light/W/Z) discrimination tasks. By integrating complementary representational strengths across diverse architectures, the framework significantly enhances feature extraction capability and generalization. Evaluated on the JetNet benchmark, the ensemble achieves superior accuracy over individual baseline models—particularly on challenging, ambiguously classified jets—demonstrating robustness and improved physical interpretability. The approach establishes a scalable deep learning paradigm for high-precision hadronic jet analysis.

0 citationsRead paper

Jet Image Generation in High Energy Physics Using Diffusion Models

Jul 31, 2025

This work pioneers the application of diffusion models to jet image generation from LHC proton-proton collision events, directly modeling the spatial distribution of particle kinematic variables in image space. We propose two class-conditional generative approaches—score-based diffusion models and consistency models—and systematically evaluate them on the JetNet dataset, which includes quark, gluon, W/Z, and top-quark jets. Compared to latent-variable paradigms, our image-space generation framework achieves both higher fidelity and improved computational efficiency. Notably, consistency models substantially outperform score-based diffusion models, yielding significant improvements in quantitative metrics such as the Fréchet Inception Distance (FID) and producing higher-quality jet images. The enhanced realism and efficiency of the generated images make this approach well-suited for high-energy physics simulation and analysis tasks, offering a promising new direction for data-driven jet modeling.

0 citationsRead paper

Recent publications

Latest Papers

Robust Prediction Variance Estimation for Gaussian Process Regression Under Covariance Smoothness Misspecification

Jun 02, 2026

This study addresses the underestimation of predictive variance in Gaussian process regression caused by misspecification of the covariance function’s smoothness, which induces bias in the mean squared prediction error (MSPE) estimation of the empirical best linear unbiased predictor (EBLUP). The authors establish, for the first time, that when the measures induced by the true and assumed covariance functions are mutually singular, the MSPE bias converges to a strictly positive limit that varies smoothly with the prediction location. Building on this insight, they propose a novel robust MSPE estimator that explicitly accounts for covariance uncertainty. Both theoretical analysis and numerical experiments demonstrate that the proposed estimator substantially outperforms four state-of-the-art alternatives across various smoothness misspecification scenarios, with its advantage becoming more pronounced as the degree of misspecification increases.

0 citationsRead paper

B-jet Tagging Using a Hybrid Edge Convolution and Transformer Architecture

Mar 22, 2026

This work proposes the Edge Convolution Transformer (ECT) model to enhance the discrimination of b-jets from c-jets and light-flavor jets in hadronic collisions. ECT uniquely integrates edge convolution with Transformer self-attention to jointly model track-level features—such as impact parameters and momentum significance—and jet-level observables, including vertex and kinematic variables. This architecture preserves local geometric structure while capturing long-range dependencies. Evaluated on ATLAS simulation data, ECT achieves an AUC of 0.9333, substantially outperforming ParticleNet (0.8904) and a pure Transformer baseline (0.9216), with notable gains in rejecting the most challenging c-jet background. Moreover, its per-jet inference latency remains below 0.060 milliseconds, satisfying the real-time triggering requirements of the LHC.

0 citationsRead paper

MEDIC: a network for monitoring data quality in collider experiments

Nov 22, 2025

In high-energy collider experiments, conventional data quality monitoring (DQM) heavily relies on real detector data, suffering from detector-specific constraints, high data acquisition costs, and extensive manual intervention—making it ill-suited for the scale and complexity of next-generation experiments. This work proposes a simulation-driven machine learning DQM framework: leveraging an enhanced Delphes fast simulator to generate controllable, labeled anomalous events; training neural networks for event-level integrity and consistency monitoring; and enabling subsystem-level fault localization. The approach decouples DQM from real experimental data, ensuring cross-detector transferability and strong scalability. Preliminary validation demonstrates that the model efficiently detects and localizes representative detector faults under simulated anomaly scenarios. This constitutes the first simulation–learning co-design paradigm for generalized, automated DQM in particle physics, paving the way for robust, scalable monitoring in future large-scale experiments.

0 citationsRead paper

Jet Image Tagging Using Deep Learning: An Ensemble Model

Aug 09, 2025

Jet substructure classification in high-energy physics is critical for discovering new physics, yet its complex, high-dimensional nature limits the performance of conventional approaches. This paper proposes a novel jet classification framework based on image-like representation and deep ensemble learning: particle-flow data are encoded as two-dimensional histograms, and a dual-branch neural network ensemble is constructed to jointly support both binary (e.g., top quark vs. light quark) and multi-class (top/light/W/Z) discrimination tasks. By integrating complementary representational strengths across diverse architectures, the framework significantly enhances feature extraction capability and generalization. Evaluated on the JetNet benchmark, the ensemble achieves superior accuracy over individual baseline models—particularly on challenging, ambiguously classified jets—demonstrating robustness and improved physical interpretability. The approach establishes a scalable deep learning paradigm for high-precision hadronic jet analysis.

0 citationsRead paper

Jet Image Generation in High Energy Physics Using Diffusion Models

Jul 31, 2025

This work pioneers the application of diffusion models to jet image generation from LHC proton-proton collision events, directly modeling the spatial distribution of particle kinematic variables in image space. We propose two class-conditional generative approaches—score-based diffusion models and consistency models—and systematically evaluate them on the JetNet dataset, which includes quark, gluon, W/Z, and top-quark jets. Compared to latent-variable paradigms, our image-space generation framework achieves both higher fidelity and improved computational efficiency. Notably, consistency models substantially outperform score-based diffusion models, yielding significant improvements in quantitative metrics such as the Fréchet Inception Distance (FID) and producing higher-quality jet images. The enhanced realism and efficiency of the generated images make this approach well-suited for high-energy physics simulation and analysis tasks, offering a promising new direction for data-driven jet modeling.

0 citationsRead paper