Selected publications

While deep learning is transforming data analysis in high-energy physics, computational challenges limit its potential. We address these challenges in the context of collider physics by introducing EveNet, an event-level foundation model pretrained on 500 million simulated collision events using a hybrid objective of self-supervised learning and physics-informed supervision. By leveraging a shared particle-cloud representation, EveNet outperforms state-of-the-art baselines across diverse tasks, including searches for heavy resonances and exotic Higgs decays, and demonstrates exceptional data efficiency in low-statistics regimes. Crucially, we validate the transferability of the model to experimental data by rediscovering the $Υ$ meson in CMS Open Data and show its capacity for precision physics through the robust extraction of quantum correlation observables stable against systematic uncertainties. These results indicate that EveNet can successfully encode the fundamental physical structure of particle interactions, which offers a unified and resource-efficient framework to accelerate discovery at current and future colliders.

Charged particle track reconstruction is a foundational task in collider experiments and the main computational bottleneck in particle reconstruction. Graph neural networks (GNNs) have shown strong performance for this problem, but costly graph construction, irregular computations, and random memory access patterns substantially limit their throughput. The recently proposed Hashing-based Efficient Point Transformer (HEPT) offers a theoretically guaranteed near-linear complexity for large point cloud processing via locality-sensitive hashing (LSH) in attention computations; however, its evaluations have largely focused on embedding quality, and the object condensation pipeline on which HEPT relies requires a post-hoc clustering step (e.g., DBScan) that can dominate runtime. In this work, we make two contributions. First, we present a unified, fair evaluation of physics tracking performance for HEPT and a representative GNN-based pipeline under the same dataset and metrics. Second, we introduce HEPTv2 by extending HEPT with a lightweight decoder that eliminates the clustering stage and directly predicts track assignments. This modification preserves HEPT's regular, hardware-friendly computations while enabling ultra-fast end-to-end inference. On the TrackML dataset, optimized HEPTv2 achieves approximately 28 ms per event on an A100 while maintaining competitive tracking efficiency. These results position HEPTv2 as a practical, scalable alternative to GNN-based pipelines for fast tracking.

A search for exotic decays of the Higgs boson $H$ into new scalar or pseudoscalar particles that subsequently decay into $b$-quarks is presented. The search considers $ZH$ production with several decay scenarios for the Higgs boson: first to a pair of identical scalars, $H\rightarrow 2a \rightarrow 4b$, second to a pair of scalars with different masses ($m_{a1}<m_{a2}$), either directly, $H\rightarrow a_1a_2 \rightarrow 4b$, or via a longer decay chain, $H \rightarrow a_1a_2 \rightarrow 3a_1 \rightarrow 6b$. The analysis uses proton-proton collision data at $\sqrt{s} = 13$ TeV collected with the ATLAS detector at the Large Hadron Collider, corresponding to an integrated luminosity of $140~\mathrm{fb}^{-1}$. No significant excess above the Standard Model prediction is observed. The search sets upper limits at 95% confidence level on the ratio of the Higgs boson production cross-section to the SM prediction times the branching ratio of Higgs bosons decaying into $4b$ or $6b$, between 4% and 25% for $σ(ZH)/σ_{\text{SM}}(ZH) \times \mathcal{B}$($H \rightarrow 2a \rightarrow 4b$), between 24% and 38% for $σ(ZH)/σ_{\text{SM}}(ZH) \times \mathcal{B}$($H \rightarrow a_1a_2 \rightarrow 4b$), and between 10% and 20% for $σ(ZH)/σ_{\text{SM}}(ZH) \times \mathcal{B}$($H \rightarrow a_1a_2 \rightarrow 3a_1 \rightarrow 6b$), depending on the masses of the scalar particles.

A search for a pseudoscalar $a$ produced in association with a top-quark pair, or in association with a single top quark plus a $W$ boson, with the pseudoscalar decaying into $b$-quarks ($a\rightarrow b\bar{b}$), is performed using the full Run 2 data sample using a dileptonic decay mode signature. The search covers pseudoscalar boson masses between 12-100 GeV and involves both the kinematic regime where the decay products of the pseudoscalar are reconstructed as two standard $b$-tagged small-radius jets, or merged into a large-radius jet due to its Lorentz boost. No significant excess relative to expectations is observed. Assuming a branching ratio BR($a\rightarrow b\bar{b}$)=100$\% $, the range of pseudoscalar masses between 50 and 80 GeV is excluded at 95$\% $ confidence level for a coupling of the pseudoscalar to the top quark of 0.5, while a coupling of 1.0 is excluded at 95$\% $ confidence level for the masses considered, with the coupling defined as the strength modifier of the Standard Model Yukawa coupling.

Scientific discoveries are often made by finding a pattern or object that was not predicted by the known rules of science. Oftentimes, these anomalous events or objects that do not conform to the norms are an indication that the rules of science governing the data are incomplete, and something new needs to be present to explain these unexpected outliers. The challenge of finding anomalies can be confounding since it requires codifying a complete knowledge of the known scientific behaviors and then projecting these known behaviors on the data to look for deviations. When utilizing machine learning, this presents a particular challenge since we require that the model not only understands scientific data perfectly but also recognizes when the data is inconsistent and out of the scope of its trained behavior. In this paper, we present three datasets aimed at developing machine learning-based anomaly detection for disparate scientific domains covering astrophysics, genomics, and polar science. We present the different datasets along with a scheme to make machine learning challenges around the three datasets findable, accessible, interoperable, and reusable (FAIR). Furthermore, we present an approach that generalizes to future machine learning challenges, enabling the possibility of large, more compute-intensive challenges that can ultimately lead to scientific discovery.

Optimizing charged-particle track reconstruction algorithms is crucial for efficient event reconstruction in Large Hadron Collider (LHC) experiments due to their significant computational demands. Existing track reconstruction algorithms have been adapted to run on massively parallel coprocessors, such as graphics processing units (GPUs), to reduce processing time. Nevertheless, challenges remain in fully harnessing the computational capacity of coprocessors in a scalable and non-disruptive manner. This paper proposes an inference-as-a-service approach for particle tracking in high energy physics experiments. To evaluate the efficacy of this approach, two distinct tracking algorithms are tested: Patatrack, a rule-based algorithm, and Exa$.$TrkX, a machine learning-based algorithm. The as-a-service implementations show enhanced GPU utilization and can process requests from multiple CPU cores concurrently without increasing per-request latency. The impact of data transfer is minimal and insignificant compared to running on local coprocessors. This approach greatly improves the computational efficiency of charged particle tracking, providing a solution to the computing challenges anticipated in the High-Luminosity LHC era.

A combination of searches for singly and doubly charged Higgs bosons, $H^{\pm}$ and $H^{\pm\pm}$, produced via vector-boson fusion is performed using 140 fb$^{-1}$ of proton-proton collisions at a centre-of-mass energy of 13 TeV, collected with the ATLAS detector during Run 2 of the Large Hadron Collider. Searches targeting decays to massive vector bosons in leptonic final states (electrons or muons) are considered. New constraints are reported on the production cross-section times branching fraction for charged Higgs boson masses between 200 GeV and 3000 GeV. The results are interpreted in the context of the Georgi-Machacek model for which the most stringent constraints to date are set for the masses considered in the combination.

This paper presents a search for exotic decays of the Higgs boson into a pair of new pseudoscalar particles, $H\rightarrow aa$, where one pseudoscalar decays into a $b$-quark pair and the other decays into a $τ$-lepton pair, in the mass range $12\leq m_{a}\leq 60$ GeV. The analysis uses $pp$ collision data at $\sqrt{s} = 13$ TeV collected with the ATLAS detector at the LHC, corresponding to an integrated luminosity of 140 ${fb}^{-1}$. No significant excess above the Standard Model (SM) prediction is observed. Assuming the SM Higgs boson production cross-section, the search sets upper limits at 95% confidence level on the branching ratio of Higgs bosons decaying into $b\bar{b}τ^+τ^-$, $\mathcal{B}(H \rightarrow aa \rightarrow b\bar{b}τ^+τ^-)$, between 2.2% and 3.9% depending on the pseudoscalar mass.

Several processes studied by the ATLAS experiment at the Large Hadron Collider produce low-momentum $b$-flavored hadrons in the final state. This paper describes the calibration of a dedicated tagging algorithm that identifies $b$-flavored hadrons outside of hadronic jets by reconstructing the soft secondary vertices originating from their decays. The calibration is based on a proton-proton collision dataset at a center-of-mass energy of 13 TeV corresponding to an integrated luminosity of 140 fb$^{-1}$. Scale factors used to correct the algorithm's performance in simulated events are extracted for the $b$-tagging efficiency and the mistag rate of the algorithm using a data sample enriched in $t\bar{t}$ events. Several orthogonal measurement regions are defined, binned as a function of the multiplicities of soft secondary vertices and jets containing a $b$-flavored hadron in the event. The mistag rate scale factors are estimated separately for events with low and high average number of interactions per bunch crossing. The results, which are derived from events with low missing transverse momentum, are successfully validated in a phase space characterized by high missing transverse momentum and therefore are applicable to new physics searches carried out in either phase space regimes.

This paper describes a search for beyond the Standard Model decays of the Higgs boson into a pair of new spin-0 particles subsequently decaying into $b$-quark pairs, $H \rightarrow aa \rightarrow (b\bar{b})(b\bar{b})$, using proton-proton collision data collected by the ATLAS detector at the Large Hadron Collider at center-of-mass energy $\sqrt{s}=13$ TeV. This search focuses on the regime where the decay products are collimated and in the range $15 \leq m_a \leq 30$ GeV and is complementary to a previous search in the same final state targeting the regime where the decay products are well separated and in the range $20 \leq m_a \leq 60$ GeV. A novel strategy for the identification of the $a \rightarrow b\bar{b}$ decays is deployed to enhance the efficiency for topologies with small separation angles. The search is performed with 36 fb$^{-1}$ of integrated luminosity collected in 2015 and 2016 and sets upper limits on the production cross-section of $H \rightarrow aa \rightarrow (b\bar{b})(b\bar{b})$, where the Higgs boson is produced in association with a $Z$ boson.

Several extensions of the Standard Model predict associated production of dark-matter particles with a Higgs boson. Such processes are searched for in final states with missing transverse momentum and a Higgs boson decaying to a $b\bar b$ pair with the ATLAS detector using 36.1 fb$^{-1}$ of $pp$ collisions at a center-of-mass energy of 13 TeV at the LHC. The observed data are in agreement with the Standard Model predictions and limits are placed on the associated production of dark-matter particles and a Higgs boson.

The increasing computational demand from growing data rates and complex machine learning (ML) algorithms in large-scale scientific experiments has driven the adoption of the Services for Optimized Network Inference on Coprocessors (SONIC) approach. SONIC accelerates ML inference by offloading it to local or remote coprocessors to optimize resource utilization. Leveraging its portability to different types of coprocessors, SONIC enhances data processing and model deployment efficiency for cutting-edge research in high energy physics (HEP) and multi-messenger astrophysics (MMA). We developed the SuperSONIC project, a scalable server infrastructure for SONIC, enabling the deployment of computationally intensive tasks to Kubernetes clusters equipped with graphics processing units (GPUs). Using NVIDIA Triton Inference Server, SuperSONIC decouples client workflows from server infrastructure, standardizing communication, optimizing throughput, load balancing, and monitoring. SuperSONIC has been successfully deployed for the CMS and ATLAS experiments at the CERN Large Hadron Collider (LHC), the IceCube Neutrino Observatory (IceCube), and the Laser Interferometer Gravitational-Wave Observatory (LIGO) and tested on Kubernetes clusters at Purdue University, the National Research Platform (NRP), and the University of Chicago. SuperSONIC addresses the challenges of the Cloud-native era by providing a reusable, configurable framework that enhances the efficiency of accelerator-based inference deployment across diverse scientific domains and industries.

This is a conference proceeding; the abstract can be found in the corresponding proceedings or public note.

The FAIR Universe HiggsML Uncertainty Challenge focused on measuring the physical properties of elementary particles with imperfect simulators. Participants were required to compute and report confidence intervals for a parameter of interest regarding the Higgs boson while accounting for various systematic (epistemic) uncertainties. The dataset is a tabular dataset of 28 features and 280 million instances. Each instance represents a simulated proton-proton collision as observed at CERN's Large Hadron Collider in Geneva, Switzerland. The features of these simulations were chosen to capture key characteristics of different types of particles. These include primary attributes, such as the energy and three-dimensional momentum of the particles, as well as derived attributes, which are calculated from the primary ones using domain-specific knowledge. Additionally, a label feature designates each instance's type of proton-proton collision, distinguishing the Higgs boson events of interest from three background sources. As outlined in this paper, the permanent release of the dataset allows long-term benchmarking of new techniques. The leading submissions, including Contrastive Normalising Flows and Density Ratios estimation through classification, are described. Our challenge has brought together the physics and machine learning communities to advance our understanding and methodologies in handling systematic uncertainties within AI techniques.

Several flavor tagging algorithms exist in ATLAS for jets containing two 𝑏-hadrons. These double-𝑏 tagger algorithms focus on high transverse-momentum jets, usually above 200 GeV. This work describes the development of a new double-𝑏 tagger for jets below 200~GeV. The algorithm relies on large radius track-jets which can be reconstructed at low transverse momenta and implements a neural network architecture based on Deep Sets that uses displaced tracks, secondary vertices, and substructure information to identify the presence of multiple 𝑏-hadrons. A measurement of the efficiency of the algorithm is performed in 𝑡𝑡ˉ and 𝑍+jets events using the collision data from the Large Hadron Collider at 𝑠 = 13 TeV center-of-mass energy recorded with the ATLAS detector between 2015 and 2018, corresponding to an integrated luminosity of 139 fb−1. Note: All figures including auxiliary figures are available at https://atlas.web.cern.ch/Atlas/GROUPS/PHYSICS/PUBNOTES/ATL-PHYS-PUB-2022-042

This is an ATLAS conference note; see the corresponding public document page for the abstract.