arXiv subjects

Aaron Wang

Publications and source records attributed to Aaron Wang.

14 recordsLinked to original sources

arXiv2026

Patch Hierarchical Attention Transformer for Efficient Particle Jet Tagging

Real-time jet tagging is critical for identifying short-lived particle decays in the high-throughput detectors of the Large Hadron Collider, where real-time trigger systems responsible for deciding which collision events to store impose strict latency and accuracy constraints. While transformer architectures achieve the highest jet tagging accuracy when compute is unconstrained, their quadratic self-attention cost makes inference restrictive on trigger budget. Existing efficient variants reduce the computational cost, but hinder the classification performance. To address this limitation, we introduce the Patch Hierarchical Attention Transformer (PHAT-JeT), which combines two mechanisms: a physics-inspired geometric message-passing module that encodes local detector-plane structure, and a hierarchical patch-based attention scheme that computes exact attention within small particle groups while preserving global context through lightweight patch-token communication. Within a restricted budget, PHAT-JeT achieves state-of-the-art accuracy and background rejection among all resource-constrained jet tagging models on four benchmarks (\textsc{hls4ml}, JetClass, Top Tagging, and Quark--Gluon). Our code is available at https://github.com/aaronw5/PHAT-JeT.

hep-ex↗

arXiv2026

SNAC-Pack 2.0: Scaled-Out Surrogate Neural Architecture Codesign

Neural architecture search (NAS) is a powerful approach for automating model design, but existing methods often optimize for accuracy alone or rely on proxy metrics such as bit operations (BOPs) that correlate poorly with hardware cost. This gap is particularly large for FPGA deployment, where cost is dominated by a multi-dimensional budget of lookup tables, DSPs, flip-flops, BRAM, and latency. We present the Surrogate Neural Architecture Codesign Package (SNAC-Pack), an open-source AutoML framework for hardware-aware neural architecture codesign and end-to-end FPGA deployment. SNAC-Pack runs a multi-objective global search with Optuna and NSGA-II, loading trials to a shared SQLite store that enables parallel workers across compute nodes. A hardware surrogate model outputs per-trial resource and latency estimates, avoiding the synthesis cost that would otherwise dominate the search loop. A local search stage then applies quantization-aware training (QAT) together with iterative magnitude pruning in a combined compression loop, after which the final model is synthesized to FPGA firmware via the hls4ml Python library. A YAML configuration and an optional agentic frontend let users run the pipeline on new datasets without modifying the framework. We demonstrate SNAC-Pack on jet classification at the Large Hadron Collider and superconducting qubit readout, discovering compact Pareto-optimal architectures that either match baseline task performance while reducing FPGA resource utilization or substantially reduce hardware utilization with only modest decreases in task performance. In the qubit readout case, SNAC-Pack also reduces the design space exploration process from months of manual fine-tuning to hours of automated search.

cs.LG↗

arXiv2025

A Model Intercomparison Study of Mixed-Phase Clouds in a Laboratory Chamber

Mixed-phase clouds, composed of supercooled liquid droplets and ice crystals, play a critical role in weather and climate systems. Their complex microphysical interactions and coupling with turbulence at microscales govern the cloud properties at macroscales, yet remain challenging to observe and quantify under atmospheric conditions. This model intercomparison study utilizes ten model configurations to simulate mixed-phase cloud evolution in the Michigan Technological University's Pi Chamber. The models span a range of frameworks, including box models, direct numerical simulation, and large-eddy simulation models, and incorporate both bin and Lagrangian microphysics. Each model was tuned to reproduce the observed liquid-phase steady state prior to ice injection. Ice particles were then introduced into the domain at various rates to examine cloud glaciation behavior. By the intercomparison design, all models successfully reproduced the observed mean droplet radius and number concentration during the liquid-phase stage. Increasing ice particle injection rates led to consistent qualitative trends across models: depletion of liquid water, reduced total water content, and a shift in particle size distributions toward larger radii. However, quantitative differences arose due to variations in model treatment in dynamics and microphysics, including subgrid-scale turbulence parameterizations, wall forcing, and particle removal parameterizations. Most models that simulate the full chamber retained liquid droplets near the lower boundary, where supersaturation forcing is strongest and droplets are replenished before mixing into the core region. These surviving liquids droplets were absent in simulations assuming a well-mixed domain, excluding the near-wall region, or using coarse grid spacing.

physics.ao-ph↗

arXiv2025

Why Is Attention Sparse In Particle Transformer?

Transformer-based models have achieved state-of-the-art performance in jet tagging at the CERN Large Hadron Collider (LHC), with the Particle Transformer (ParT) representing a leading example of such models. A striking feature of ParT is its sparse, nearly binary, attention structure, raising questions about the origin of this behavior and whether it encodes physically meaningful correlations. In this work, we investigate the source of ParT's sparse attention by comparing models trained on multiple benchmark datasets and examine the relative contributions of the attention term and the physics-inspired interaction matrix before softmax. We find that binary sparsity arises primarily from the attention mechanism itself, with the interaction matrix playing a secondary role. Moreove, we show that ParT is able to identify key jet substructure elements, such as leptons in semileptonic top decays, even without explicit particle identification inputs. These results provide new insight into the interpretability of transformer-based jet taggers and clarify the conditions under which sparse attention patterns emerge in ParT.

hep-ph↗

arXiv2025

Cloud droplet size distribution and optical properties only weakly linked to aerosol size

Changes in aerosol concentrations can modify cloud brightness, producing a strong but poorly constrained influence on Earth's energy balance. Because cloud reflectivity depends on the size distribution of cloud droplets, and aerosol size strongly governs activation into droplets, one might expect cloud properties to be sensitive to aerosol size distributions. Here we show, through a combination of cloud chamber experiments and high-resolution simulations, that cloud microphysical and optical properties are often insensitive to aerosol size. Detectable impacts on cloud optical properties occur only under weak convective forcing and high aerosol concentrations. These results indicate that, in most conditions, cloud reflectivity can be predicted from aerosol number alone without detailed knowledge of aerosol size distributions, providing new constraints on how aerosol perturbations affect climate.

physics.ao-ph↗

arXiv2025

Spatially Aware Linear Transformer (SAL-T) for Particle Jet Tagging

Transformers are very effective in capturing both global and local correlations within high-energy particle collisions, but they present deployment challenges in high-data-throughput environments, such as the CERN LHC. The quadratic complexity of transformer models demands substantial resources and increases latency during inference. In order to address these issues, we introduce the Spatially Aware Linear Transformer (SAL-T), a physics-inspired enhancement of the linformer architecture that maintains linear attention. Our method incorporates spatially aware partitioning of particles based on kinematic features, thereby computing attention between regions of physical significance. Additionally, we employ convolutional layers to capture local correlations, informed by insights from jet physics. In addition to outperforming the standard linformer in jet classification tasks, SAL-T also achieves classification results comparable to full-attention transformers, while using considerably fewer resources with lower latency during inference. Experiments on a generic point cloud classification dataset (ModelNet10) further confirm this trend. Our code is available at https://github.com/aaronw5/SAL-T4HEP.

cs.LG↗

arXiv2024

Interpreting Transformers for Jet Tagging

Machine learning (ML) algorithms, particularly attention-based transformer models, have become indispensable for analyzing the vast data generated by particle physics experiments like ATLAS and CMS at the CERN LHC. Particle Transformer (ParT), a state-of-the-art model, leverages particle-level attention to improve jet-tagging tasks, which are critical for identifying particles resulting from proton collisions. This study focuses on interpreting ParT by analyzing attention heat maps and particle-pair correlations on the $\eta$-$\phi$ plane, revealing a binary attention pattern where each particle attends to at most one other particle. At the same time, we observe that ParT shows varying focus on important particles and subjets depending on decay, indicating that the model learns traditional jet substructure observables. These insights enhance our understanding of the model's internal workings and learning process, offering potential avenues for improving the efficiency of transformer architectures in future high-energy physics applications.

hep-ph↗

arXiv2023

Fermi Surface Evolution and Anomalous Hall Effect in an Ideal Type-II Weyl Semimetal

Weyl semimetals (WSMs) are three-dimensional topological materials that exhibit fascinating properties due to the presence of Weyl nodes in their band structure. However, existing WSMs discovered so far often possess multiple pairs of Weyl nodes, posing a challenge in disentangling the contributions to transport phenomena from different energy bands. To overcome this challenge, we have identified field-induced ferromagnetic MnBi$_{2-x}$Sb$_{x}$Te$_{4}$ as an ideal type-II WSM with a single pair of Weyl nodes. By employing a combination of quantum oscillations and high-field Hall measurements, we have resolved the evolution of Fermi-surface sections as the Fermi level is tuned across the charge neutrality point, precisely matching the band structure of an ideal type-II WSM. Furthermore, the anomalous Hall conductivity exhibits a heartbeat-like behavior as the Fermi level is tuned across the Weyl nodes, a unique feature previously predicted for a type-II WSM. Our findings establish MnBi$_{2-x}$Sb$_{x}$Te$_{4}$ as an ideal platform for further investigation into Weyl physics.

cond-mat.mes-hall↗

arXiv2022

Machine Learning Prediction of COVID-19 Severity Levels From Salivaomics Data

The clinical spectrum of severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2), the strain of coronavirus that caused the COVID-19 pandemic, is broad, extending from asymptomatic infection to severe immunopulmonary reactions that, if not categorized properly, may be life-threatening. Researchers rate COVID-19 patients on a scale from 1 to 8 according to the severity level of COVID-19, 1 being healthy and 8 being extremely sick, based on a multitude of factors including number of clinic visits, days since the first sign of symptoms, and more. However, there are two issues with the current state of severity level designation. Firstly, there exists variation among researchers in determining these patient scores, which may lead to improper treatment. Secondly, researchers use a variety of metrics to determine patient severity level, including metrics involving plasma collection that require invasive procedures. This project aims to remedy both issues by introducing a machine learning framework that unifies severity level designations based on noninvasive saliva biomarkers. Our results show that we can successfully use machine learning on salivaomics data to predict the severity level of COVID-19 patients, indicating the presence of viral load using saliva biomarkers.

q-bio.QM↗

arXiv2022

Deep Learning Methods for Protein Family Classification on PDB Sequencing Data

Composed of amino acid chains that influence how they fold and thus dictating their function and features, proteins are a class of macromolecules that play a central role in major biological processes and are required for the structure, function, and regulation of the body's tissues. Understanding protein functions is vital to the development of therapeutics and precision medicine, and hence the ability to classify proteins and their functions based on measurable features is crucial; indeed, the automatic inference of a protein's properties from its sequence of amino acids, known as its primary structure, remains an important open problem within the field of bioinformatics, especially given the recent advancements in sequencing technologies and the extensive number of known but uncategorized proteins with unknown properties. In this work, we demonstrate and compare the performance of several deep learning frameworks, including novel bi-directional LSTM and convolutional models, on widely available sequencing data from the Protein Data Bank (PDB) of the Research Collaboratory for Structural Bioinformatics (RCSB), as well as benchmark this performance against classical machine learning approaches, including k-nearest neighbors and multinomial regression classifiers, trained on experimental data. Our results show that our deep learning models deliver superior performance to classical machine learning methods, with the convolutional architecture providing the most impressive inference performance.

q-bio.QM↗

arXiv2022

Ultra-low latency recurrent neural network inference on FPGAs for physics applications with hls4ml

Recurrent neural networks have been shown to be effective architectures for many tasks in high energy physics, and thus have been widely adopted. Their use in low-latency environments has, however, been limited as a result of the difficulties of implementing recurrent architectures on field-programmable gate arrays (FPGAs). In this paper we present an implementation of two types of recurrent neural network layers -- long short-term memory and gated recurrent unit -- within the hls4ml framework. We demonstrate that our implementation is capable of producing effective designs for both small and large models, and can be customized to meet specific design requirements for inference latencies and FPGA resources. We show the performance and synthesized designs for multiple neural networks, many of which are trained specifically for jet identification tasks at the CERN Large Hadron Collider.

cs.LG↗

arXiv2020

SimpleChrome: Encoding of Combinatorial Effects for Predicting Gene Expression

Due to recent breakthroughs in state-of-the-art DNA sequencing technology, genomics data sets have become ubiquitous. The emergence of large-scale data sets provides great opportunities for better understanding of genomics, especially gene regulation. Although each cell in the human body contains the same set of DNA information, gene expression controls the functions of these cells by either turning genes on or off, known as gene expression levels. There are two important factors that control the expression level of each gene: (1) Gene regulation such as histone modifications can directly regulate gene expression. (2) Neighboring genes that are functionally related to or interact with each other that can also affect gene expression level. Previous efforts have tried to address the former using Attention-based model. However, addressing the second problem requires the incorporation of all potentially related gene information into the model. Though modern machine learning and deep learning models have been able to capture gene expression signals when applied to moderately sized data, they have struggled to recover the underlying signals of the data due to the nature of the data's higher dimensionality. To remedy this issue, we present SimpleChrome, a deep learning model that learns the latent histone modification representations of genes. The features learned from the model allow us to better understand the combinatorial effects of cross-gene interactions and direct gene regulation on the target gene expression. The results of this paper show outstanding improvements on the predictive capabilities of downstream models and greatly relaxes the need for a large data set to learn a robust, generalized neural network. These results have immediate downstream effects in epigenomics research and drug development.

q-bio.GN↗

arXiv2020

Small Energy Gap Revealed in CrBr3 by Scanning Tunneling Spectroscopy

CrBr$_{3}$ is a layered van der Waals material with magnetic ordering down to the 2D limit. For decades, based on optical measurements, it is believed that the energy gap of CrBr$_{3}$ is in the range of 1.68-2.1 eV. However, controversial results have indicated that the band gap of CrBr$_{3}$ is possibly smaller than that. An unambiguous determination of the energy gap is critical to the correct interpretations of the experimental results of CrBr$_{3}$. Here, we present the scanning tunneling microscopy and spectroscopy (STM/S) results of CrBr$_{3}$ thin and thick flakes exfoliated onto pyropytic graphite (HOPG) surfaces and density functional theory (DFT) calculations to reveal the small energy gap (peak-to-peak energy gap to be 0.57 eV $\pm$ 0.04 eV; or the onset signal energy gap to be 0.29 $\pm$ 0.05 eV from dI/dV spectra). Atomic resolution topography images show the defect-free crystal structure and the dI/dV spectra exhibit multiple peak features measured at 77 K. The conduction band - valence band peak pairs in the multi-peak dI/dV spectrum agree very well with all reported optical transitions. STM topography images of mono- and bi-layer CrBr$_{3}$ flakes exhibit edge degradation due to short air exposure (~15 min) during sample transfer. The unambiguously determined small energy gap settles the controversy and is the key in better understanding CrBr$_{3}$ and similar materials.

cond-mat.mtrl-sci↗

arXiv2019

Elongated Nano Domains and Molecular Intermixing induced Doping in Organic Photovoltaic Active Layers with Electric Field Treatment

The effects of the electric-field-assisted annealing on the bulk heterojunction nano-morphology in the P3HT/PCBM active layer of the organic photovoltaic cells (OPVCs) are presented here. It was widely accepted that the electric-field-assisted annealing will facilitate the P3HT, the polar polymer, to be better crystalline to enhance the charge mobility, hence the improvement of the OPVC performance. The influences on the nano-morphology of the electron donor and accepter domains are not well understood. Here, using the cross-sectional scanning tunneling microscopy and spectroscopy (XSTM/S), the electric-field-assisted annealing treatment is found to influence the molecular domains to be elongated with the orientation near the direction of the external electric field. The elongation of the molecular domains is believed to facilitate the domain percolation, which causes higher charge mobility, hence the higher short-circuit current density (Jsc). On the other hand, it was also observed that the electronic properties of the P3HT-rich and PCBM-rich domains in the electric-field-assisted annealed samples showed smaller energy band gaps and smaller molecular orbital offset between the two domains, which is argued to decrease the open circuit voltage (Voc) and negatively impact the OPVC performance. Based on the X-ray diffraction (XRD) and small angle X-ray scattering (SAXS) results, the altered electronic properties are argued to be due to the molecular intermixing induced doping effects. These results point out competing factors affecting the OPVC performance with the electric-field-assisted annealing treatment.

physics.app-ph↗