-
31/08/2026, 13:50Invited Presentation
-
Fred Chong (University of Chicago)31/08/2026, 14:00
-
Jason Mohoney (Massachusetts Inst. of Technology (US))31/08/2026, 14:30
-
Maciej Mikolaj Glowacki (CERN)
-
Stephen Neuendorffer (AMD)
-
Joey McClenaghan (General Atomics)
-
David Miller (University of Chicago (US))
-
Kai YamaguchiPoster
The LHC collides protons at a rate of 40 million collisions per second. To filter the massive amount of data for interesting physics, the real-time trigger systems inside detectors at the LHC necessitate smart and sophisticated triggers that are 1) efficient enough to simultaneously reject large backgrounds and keep enough signal, 2) compact enough to meet hardware constraints, and 3) fast...
Go to contribution page -
Gian Michele Innocenti (Massachusetts Inst. of Technology (US))Presentation
On-detector data reduction has become a primary constraint for next-generation monolithic active pixel sensors (MAPS) operated in continuous readout, which are used in particle physics to realize high-resolution, low-material-budget tracking and vertexing detectors. As pixel pitches decrease and hit rates increase, the bandwidth, power, and material-budget cost of this data transport is...
Go to contribution page -
Robert Sneiderman (Independent)Poster
Efficient-attention transformers trade direct context access for compute, and a family of memory mechanisms is offered to recover what restricted attention drops. Reported gains are hard to trust: papers vary the mechanism, the backbone, the optimizer, the token budget, and the scale all at once, and they report a single aggregate perplexity that averages over positions where local attention...
Go to contribution page -
Ruthwik Reddy SunketaPoster
Small neural networks are an important class of FPGA workload. Many scientific and edge applications run small models under tight latency budgets, including high-energy-physics triggering, network-intrusion detection, and keyword spotting. When the model is small enough to keep on-chip, FPGAs can deliver low-latency, real-time inference. Two widely-used toolchains approach deployment...
Go to contribution page -
Aleesha Kallil Tharayil (Carnegie-Mellon University (US))Poster
Accurate reconstruction of the missing transverse momentum (MET) is essential for a broad range of CMS analyses, including searches for new physics and precision measurements of the W boson mass. However, its resolution degrades significantly in high-pileup environments and is expected to degrade further at the High-Luminosity LHC. A state-of-the-art deep neural network-based MET...
Go to contribution page -
Noah Clarke Hall (CERN), Ramya Gurunathan, Vila Andela Petrovic (UCL)Presentation
High Energy Physics experiments at the energy and intensity frontier face O(Tbps) data rates that must be processed with microsecond latency. To handle the large amount of data, a two-stage selection strategy is typically deployed to select and record the most interesting events in real time, where the first stage is based on custom electronics and the second one on a heterogeneous CPU/GPU...
Go to contribution page -
Gia-Wei Chern (University of Virginia)Poster
Machine learning is rapidly transforming computational science by replacing expensive first-principles calculations with accurate, scalable surrogate models. A major challenge, however, is incorporating the fundamental physical symmetries that govern scientific simulations. We present a gauge-equivariant graph neural network (GNN) that enables scalable machine learning for lattice gauge...
Go to contribution page -
Yuheon Joh (University of California San Diego)Presentation
While transformer-based models such as the Particle Transformer have achieved strong performance offline, deploying the full model in a low-latency, resource-constrained trigger system remains an open problem. We extend the previous implementation of a quantized, integer-only transformer for jet tagging on the AMD Versal AI Engine by adding support for LayerNorm and GELU and implementing the...
Go to contribution page -
Dr Javier Hernandez Nicolau (San Diego Supercomputer Center, UC San Diego), Dr Madhusudan Gujral (San Diego Supercomputer Center, UC San Diego), Dr Mahidhar Tatineni (San Diego Supercomputer Center, UC San Diego)Tutorial
This tutorial will present the architecture, Kubernetes based systems setup, user software environment, scalability studies, and fine tuning on the Voyager system. Voyager is an US National Science Foundation funded AI-focused hardware based supercomputer. It is built using the Intel/Habana Gaudi processors (Gaudi1 and Gaudi2), has a 400 GbE interconnect from Arista for scale out training and...
Go to contribution page -
Chang Sun (California Institute of Technology (US))Presentation
Low-latency machine-learning and data-processing pipelines often contain static dataflow kernels that require fine-grained bitwidth control, arithmetic optimization, and fast hardware performance estimation. This works introduce Alkaid, a free and open-source domain-specific compiler that translates sub-microsecond latency dataflow kernels into platform-agnostic Register-Transfer Level (RTL)...
Go to contribution page -
Pritam Singh (Wichita State University)Presentation
Determining the internal physical structures of evolved stars remains a central challenge in stellar astrophysics. Standard procedures involve combining 1D stellar evolution tracks from codes like MESA with linear perturbation data from GYRE to reproduce observed stellar oscillation spectra. Although these numerical techniques are highly accurate, mapping observable surface frequencies back to...
Go to contribution page -
Elias Leutgeb (CERN), Fabian Helmberger (Vienna University of Technology (AT)), Thomas Owen James (CERN)Presentation
Deep learning techniques are being used for event classification, reconstruction, and triggering in the first stages of data selection and processing at collider experiments, such as CMS and ATLAS at the LHC. Thus far, the networks being used for low-latency inference are predominately based on fully-connected architectures. On the other hand, transformer-based ML models have demonstrated...
Go to contribution page -
Adrian Chitan (IFIN-HH (RO))Poster
Final states containing isolated electrons and photons (e/γ) have played a central
Go to contribution page
role in physics discoveries at the Large Hadron Collider (LHC) and will remain
vital to the ATLAS trigger strategy throughout the High-Luminosity LHC (HL-LHC)
programme. However, the substantially increased pile-up expected at the HL-LHC
will make it increasingly challenging to preserve high trigger... -
Yichao Lin (The George Washington University)Poster
Over the past 25 years, over 2 million X-ray sources have been serendipitously discovered by various X-ray observatories, where the majority remain unclassified. Traditional manual classification methods alone are increasingly unable to keep up with the growth of data. We present the results and lessons learned from applying a random forest classifier to X-ray sources from the XMM-Newton...
Go to contribution page -
Lauri Antti Olavi Laatu (Imperial College (GB))Presentation
Autoregressive generative models form the basis of modern LLMs. These models utilize a transformer architecture which requires caching the previous attention matrix (KV-Cache) for efficient inference. FPGAs are uniquely capable of high throughput due to their on-chip memory for both weights and biases as well as for KV-Cache. We present a GPT based autoregressive model trained using High...
Go to contribution page -
Dorian Sloot (Austrian Academy of Sciences (AT))Poster
Low-latency machine learning is essential for real-time decision systems in high-energy physics, where strict resource and latency constraints limit deployable model complexity. Aggressive quantization can potentially reduce memory and arithmetic requirements, but its effect on predictive performance must be understood before deployment. We present a controlled comparison of low-precision...
Go to contribution page -
Nhan Tran (Fermi National Accelerator Lab. (US))Topical session
Scientific AI has made tremendous progress in accelerating inference, compressing models, and deploying machine learning at the edge. However, most scientific systems remain fundamentally open-loop: AI analyzes data or makes predictions, while scientists remain responsible for interpreting results, recalibrating instruments, updating models, and adjusting control strategies. The next frontier...
Go to contribution page -
Atul GargPoster
Distributed Machine Learning (ML) systems and real-time inference pipelines are traditionally designed and benchmarked under the assumption of homogeneous, dedicated cloud infrastructure. However, in modern corporate and sovereign cloud environments, distributed ML workloads must routinely co-exist alongside long-running, memory-heavy enterprise resource planning (ERP) databases and...
Go to contribution page -
76. Capturing Neutrinos and New Physics at ProtoDUNE with a Machine Learning-Based Trigger AlgorithmCiaran Hasnip (CERN)Presentation
The Deep Underground Neutrino Experiment (DUNE) will deploy massive liquid argon time projection chamber (LArTPC) detectors to measure the properties of neutrinos with unprecedented precision. Currently, two full-scale prototypes of the DUNE far detectors (FDs) are at CERN, which are called the ProtoDUNE detectors. Recent studies indicate that these prototypes could have the potential to...
Go to contribution page -
Jim Brooke (University of Bristol (GB))Poster
We describe a CNN based approach to full event classification, for fast first level event selection at the HL-LHC. We perform hardware-aware optimisation of the network architecture, and evaluate physics performance using simulated data. This allowed a range of network models to be identified that fit within target FPGA resources and latency requirements of HL-LHC trigger systems. A candidate...
Go to contribution page -
Benedikt Maier (Imperial College (GB))Presentation
Machine learning is increasingly central to both offline data analysis and applications on edge systems at the Large
Go to contribution page
Hadron Collider (LHC). However, progress is hampered by a lack of standardised benchmarks that simultaneously address
the diverse requirements of high-energy physics, from high-precision offline inference to ultra-low-latency inference on
FPGA hardware. This paper introduces... -
Javier Hernandez-Nicolau (San Diego Supercomputer Center)Poster
The integration of AI into cosmological research is poised to significantly impact major experiments such as Simons Observatory, LiteBIRD, and CMB-S4. These projects aim to achieve unprecedented precision in mapping the cosmic microwave background (CMB), necessitating high-resolution simulations to interpret the data accurately. AI techniques, particularly those enhancing low-resolution...
Go to contribution page -
Qibin Liu (SLAC National Accelerator Laboratory (US))Poster
Recent advances in machine learning and microelectronics are enabling efficient real-time, on-chip data processing under stringent latency, power, and bandwidth constraints. Compact machine-learning models implemented directly in hardware can replace or augment fixed logic for intelligent feature extraction, classification, and denoising at the detector front-end. These capabilities are...
Go to contribution page -
Zhenghua Ma (University of California San Diego)Presentation
Extreme-edge scientific applications use machine learning models to analyze sensor data and make real-time decisions.
Go to contribution page
Their stringent latency and throughput requirements demand small batch sizes and require that model weights remain fully on-chip.
Spatial dataflow implementations are common for extreme-edge applications. Spatial dataflow works well for small networks, but it fails to scale... -
Kyungseop Yoon (Massachusetts Institute of Technology)Poster
Fast and accurate parameter estimation of binary neutron star (BNS) mergers, gravitational wave events with electromagnetic counterparts, remains a central challenge in multimessenger astronomy. Building on state space models (SSMs), we directly regress BNS merger source parameters from raw gravitational wave time series, without sampling-based inference. As a first demonstration, we focus on...
Go to contribution page -
Noah Clarke Hall (CERN), Tara TahseenPresentation
Normalizing flows are principled anomaly detectors, selecting anomalies using a probabilistic per-event likelihood. Extreme latency and resource constraints have prevented the deployment of flow likelihoods within the hardware triggers at the Large Hadron Collider. We bypass these limitations by distilling the likelihood from a large normalizing flow into lightweight student estimators...
Go to contribution page -
Paul Devouge (Université Paris-Saclay (FR))Presentation
In modern collider experiments, rare and interesting physics processes must be identified in real time while coping with enormous data rates from a large number of detector sensors. Reconstruction algorithms and trigger decisions already need to be made within microseconds, reducing the data volume by orders of magnitude with very low latency. At next generation particle accelerators such as...
Go to contribution page -
Alan XiaPresentation
Transformer-based jet taggers achieve state-of-the-art classification accuracy at the LHC but remain too computationally expensive for the FPGA-based Level-1 trigger systems planned for the HL-LHC upgrade. We present two complementary attention architectures, SAL-T and PHAT-JeT, that close this gap by introducing physics-motivated structural priors into the attention mechanism while preserving...
Go to contribution page -
Vaibhav LohiaPoster
Real-time identification of jets from boosted heavy particles is vital for the High-Luminosity LHC physics program at CMS. However, deploying highly expressive architectures like the Particle Transformer within the hardware-constrained Level-1 (L1) trigger is severely limited by the quadratic scaling $\mathcal{O}(N^2)$ of standard self-attention. This computational burden makes...
Go to contribution page -
Sabrina Giorgetti (CERN)Presentation
During Run 3 of the LHC, the CMS experiment introduced real-time anomaly detection algorithms for online event selection for the first time. The deployment of AXOL1TL and CICADA demonstrated the feasibility of signal-agnostic autoencoder models for ultra-low-latency inference in Level-1 trigger hardware. Building on AXOL1TL, a second anomaly-detection layer has been explored for the...
Go to contribution page -
Andrew Whitbeck (Fermi National Accelerator Lab. (US))Presentation
We present a surrogate-based learning framework for optimizing control policies in two particle accelerator applications: Lorentz Force Detuning (LFD) compensation in superconducting radio-frequency (SRF) cavities, and spill intensity regulation for the Mu2e experiment at Fermilab. In both systems, physical dynamics are complex and nonlinear, motivating data-efficient surrogate models that...
Go to contribution page -
Melissa Medina PeregrinaPoster
A next-generation neutrinoless double beta decay (0νββ) search in ¹³⁶Xe has the potential to uncover lepton number violation, and determine if neutrinos are their own antiparticle. This rare decay, if discovered, would demonstrate Physics beyond the Standard Model and provide key insights into the evolution of the universe.
Fast machine learning allows real time data processing in hardware,...
Go to contribution page -
Raymond DuenasPoster
In high radiation environments, hardware accelerators are prone to radiation-induced bit flips, leading to data corruption. For example, scientists at the Large Hadron Collider (LHC) seek to deploy hardware-accelerated neural networks in environments with radiation 1000X higher than that seen in space [1]. Ensuring reliable data collection, such as at the LHC, requires developing hardware that...
Go to contribution page -
Zhiqiang Que (University of Bristol), Chang Sun (California Institute of Technology (US))Poster
Boosted decision trees (BDTs) are widely used in latency-critical applications, but efficient FPGA deployment remains challenging. Existing designs often rely on uniform or manually tuned fixed-point formats, which can introduce unnecessary hardware cost or accuracy loss. This work presents FQTree, a framework combining fine-grained quantization-aware training with automatic hardware...
Go to contribution page -
Marius Köppel (ETH Zurich (CH))Poster
Quantization‑aware training (QAT) is essential for delivering low‑latency inference on the FPGAs that power high‑energy‑physics (HEP) experiments. QKeras has become the de‑facto Keras‑based framework for QAT in this community and is tightly coupled with the hls4ml toolchain, which converts Keras models into synthesizable HLS code. The transition from Keras 2 to Keras 3 introduced a new "ops"...
Go to contribution page -
Liv Helen Vage (Princeton University (US))Presentation
Machine learning models are increasingly deployed where knowing when to trust them matters, yet uncertainty quantification (UQ) is often avoided since it can be costly to run. We show that this cost reduces dramatically for fast architectures which reduce to binary computations, such as logic neural networks. Replacing floating-point arithmetic with boolean gates makes inference fast and...
Go to contribution page -
Adrian Chitan (IFIN-HH (RO))Presentation
The absence of beyond-Standard-Model physics discoveries at the LHC suggests that new physics may evade conventional trigger strategies. Existing ATLAS triggers control data collection rates either by applying energy thresholds or by targeting event topologies specific to particular models. Unsupervised machine learning offers a compelling alternative via anomaly detection, enabling a...
Go to contribution page -
Sandip Roy (University of California San Diego)Poster
Cosmological simulations are too computationally expensive to support exhaustive scans over dark matter parameter space. Generative models, including diffusion models and flow models, offer a promising route to simulation emulation using existing suites of numerical simulations. In this talk, I will describe diffusion and flow-based generative models as transport processes: they evolve...
Go to contribution page -
Mr Venkata Sai Prathyush Turaga (Texas Tech University)Poster
Configuring hls4ml(High-Level Synthesis for Machine Learning) for a new scientific task is a manual loop. The user sets precision, reuse factor, strategy, IO type, and backend, runs HLS synthesis, and reads back latency and resource usage. A single synthesis pass takes minutes to hours, and the search space grows quickly when settings are tuned per layer. Domain experts in particle physics,...
Go to contribution page -
Yuan-Tang Chou (University of Washington (US)), Jan-Frederik Schulte (Purdue University (US))Presentation
Charged-particle tracking is a core reconstruction task in high-energy physics that directly impacts all reconstruction and physics studies. At the High-Luminosity Large Hadron Collider (HL-LHC), this must be done under much higher pile-up while preserving both tracking quality and computational efficiency. While existing graph-based approaches achieve physics performance comparable to...
Go to contribution page -
Benjamin Ramhorst (Imperial College (GB)), Jan-Frederik Schulte (Purdue University (US))Topical session
[
Go to contribution pagehls4ml][1] is a flexible, open-source platform that addresses these hardware development challenges by converting high-level neural networks from PyTorch, Keras, or similar frameworks into high-level synthesis (HLS) kernels suitable for deployment on FPGAs and ASICs. During this topical session, developers and users will discuss the evolving roadmap, open issues and challenges,... -
Georgios Flengas (CERN), Marius Köppel (ETH Zurich (CH))Tutorial
For the deployment of machine learning (ML) models with strict requirements on latency and computing resources, field-programmable gate arrays (FPGAs) have emerged as a preferred hardware platform, as they offer low-level hardware control and easy reprogrammability. However, deploying models on these devices requires expert knowledge of dedicated programming techniques, such as high-level...
Go to contribution page -
Adrian Chitan (IFIN-HH (RO))Presentation
Particle tracking at Large Hadron Collider (LHC) experiments is a crucial component of particle reconstruction, yet it remains one of the most computationally challenging tasks in this process. As we approach the High-Luminosity LHC (HL-LHC) era, the complexity of tracking is expected to increase significantly. Leveraging coprocessors such as GPUs presents a promising solution to the rising...
Go to contribution page -
Brock Sorenson (Arizona State University)Presentation
IRONSmith: A Visual Dataflow Design Environment for AMD Ryzen AI NPUs
Brock Sorenson, Samer Ali, Curt Bansil, Aman Arora — Arizona State University
Machine learning inference increasingly relies on specialized hardware
Go to contribution page
accelerators for throughput and power efficiency. Neural Processing Units
(NPUs), such as the AMD Ryzen AI NPU, offer significant ML advantages over
CPUs... -
Kalib McEuen (Univ. of Illinois Chicago (US))Poster
The Large Hadron Collider (LHC) collides protons at a rate of 40 million collisions per second. To filter the massive amount of data for interesting physics, the real-time trigger systems inside detectors at the LHC necessitate smart and sophisticated triggers that are 1) efficient enough to simultaneously reject large backgrounds and keep enough signal, 2) compact enough to meet hardware...
Go to contribution page -
Eric Anton Moreno (Massachusetts Institute of Technology (US))Presentation
Experimental high energy physics analysis is traditionally a multi-year effort dominated by repetitive code development that demands little physics insight. We show that LLM-based AI agents can already autonomously execute substantial portions of this pipeline, collapsing the implementation bottleneck to roughly ten hours of wall-clock time. We present Just Furnish Context (JFC), a...
Go to contribution page -
Romina Soledad MolinaTutorial
Abstract
Deploying neural networks on FPGAs remains a significant barrier in embedded AI. Exploring the design space across compression strategies (pruning, quantization, knowledge distillation) and hardware configurations is often fragmented, requiring deep expertise across multiple toolchains.This tutorial introduces KalEdge, a hardware-aware platform unifying the ML-to-FPGA...
Go to contribution page -
Noah Paladino (Massachusetts Inst. of Technology (US))Presentation
As low-latency machine learning techniques are adopted in data collection and processing across various domains, a standard hardware platform becomes necessary to ease adoption. While many FPGA manufacturers provide a few development kits to demonstrate device capabilities, most of these lack the raw inputs necessary to cost-effectively build custom data acquisition systems. We present KrIO, a...
Go to contribution page -
Erdem Yigit Ertorer (Carnegie-Mellon University (US))Poster
Transformers have gained significant traction in high-energy physics (HEP) experiments. However, many HEP applications require ultra-fast processing and the quadratic complexity of standard self-attention becomes prohibitive. Fortunately, there are several approaches to address this bottleneck. One is to replace this standard self-attention with linearized attention. In this work, we explore...
Go to contribution page -
Eric Anton Moreno (Massachusetts Institute of Technology (US)), Nan Huang (UC San Diego), Yue MaTopical session
LLM agents are reshaping how particle physicists interact with data, code, and the research process itself. From "vibe coding" to autonomous scientist frameworks capable of hypothesis generation and experiment design, these tools are beginning to transform workflows in high-energy physics.
This BoF session invites practitioners, skeptics, and enthusiasts to share experiences and openly...
Go to contribution page -
Dr Jia Wang (NOAA Great Lakes Environmental Research Laboratory)Presentation
Accurate Great Lakes ice forecasting is essential for maritime operations, coastal protection, and offshore engineering, yet traditional physics-based models like FVCOM–CICE remain computationally expensive. This study develops a deep neural network surrogate trained on an extensive FVCOM–CICE hindcast using ERA5 2-m air temperature and FVCOM surface water temperature as physically motivated...
Go to contribution page -
Maryam Bayat Makou (Southern Methodist University (US))Poster
Efficient identification of boosted hadronic objects is an important challenge for the ATLAS Phase-II trigger system. This work investigates machine-learning-based large-R jet tagging for the ATLAS Level-0 Global Trigger using trigger-level calorimeter information under HL-LHC pile-up conditions.
Go to contribution page
Using simulated $HH\rightarrow b\bar{b}b\bar{b}$ signal and QCD multijet background events, we... -
Zepeng Li (University of Hawaii at Manoa)Poster
The COHERENT experiment has demonstrated coherent elastic neutrino–nucleus scattering (CEvNS) at the Spallation Neutron Source, establishing a powerful neutral-current channel for probing all neutrino flavors. Tonne-scale CEvNS detectors, especially the cryogenic CsI detector, offer a promising opportunity to detect neutrinos from a Galactic core-collapse supernova. A key challenge is that the...
Go to contribution page -
Matteo Marchegiani (Carnegie-Mellon University (US))Presentation
The calorimeters employed in large experiments at the Phase-1 LHC were designed to cope with the radiation levels and hit multiplicities expected in proton-proton collisions with an average of 20–50 interactions per bunch crossing. However, the High-Luminosity LHC (HL-LHC) will operate with pileup levels of up to 200 interactions per crossing. The radiation environment is especially...
Go to contribution page -
Jędrzej MaczanPoster
Activation checkpointing minimizes the runtime of neural networks under a given memory budget, by selecting which intermediate tensors to store and which to recompute. PyTorch solves this as a 0/1 knapsack problem, where operations from a joint forward-backward computation graph are items with a memory cost (weight) and a runtime saving (value). The default solver, dp_knapsack, allocates a...
Go to contribution page -
Tae Min Hong (University of Pittsburgh (US))Presentation
We present an implementation of edge AI to compress data on an in-memory analog content-addressable memory (ACAM) device. A variational autoencoder is trained on a simulated sample of energy measurements from incident high-energy electrons on a generic three-layer scintillator-based calorimeter. The encoding part is distilled into tabular format by regressing the latent space variables using...
Go to contribution page -
Diego Osvaldo Ochoa de la CruzPoster
Post-Asymptotic Giant Branch (post-AGB) stars are critical, short-lived transition objects in stellar evolution. However, only 394 confirmed post-AGB stars are currently known, heavily limiting our ability to constrain AGB models. While recent massive all-sky surveys (e.g., 2MASS, WISE, SDSS, IGAPS, and VPHAS+) offer an unprecedented opportunity to discover new candidates at scale, exploiting...
Go to contribution page -
Jack RedepenningPoster
Detection of gravitational waves (GWs) has opened new roads in exploring and analyzing astrophysical data. Not only can we learn more about gravitational waves themselves, but this also allows us to perform multi-messenger astronomy, detecting both the GW and electromagnetic (EM) signals. GW170817 demonstrated the power of multi-messenger detections. It confirmed that neutron star mergers...
Go to contribution page -
Ethan Colbert (Purdue University (US)), Noah Paladino (Massachusetts Inst. of Technology (US))Poster
As scientific data analysis workflows embrace machine learning, demand for elastic GPU compute is rising. Collaborations have begun to turn to inference-as-a-service solutions, which rely on inference servers like NVIDIA Triton, to meet these needs. However, inference servers designed for the cloud have notable shortcomings when deployed within an HPC environment, including difficulties...
Go to contribution page -
David Jiang (Univ. Illinois at Urbana Champaign (US))Presentation
Modern scientific instruments operate under increasingly extreme constraints on bandwidth, latency, and power. Inference at the sensor edge determines experimental data collection efficiency by deciding which information to save for further analysis. Particle tracking detectors at the Large Hadron Collider exemplify this challenge: pixelated silicon sensors generate rich spatiotemporal...
Go to contribution page -
Zepeng Li (University of Hawaii at Manoa)
-
Rotem Dover (Department of Particle Physics and Astrophysics, Weizmann Institute of Science)Presentation
High-fidelity simulation of particle–matter interactions provides the essential theoretical reference for diverse physics disciplines, yet generating synthetic datasets at the scale of current and future experiments has become prohibitive. Here, we introduce PHIN-GAN, a novel physics-informed generative adversarial network designed to address this challenge. We derive a set of analytical...
Go to contribution page -
FEMI JOHNSONPoster
The accuracy of machine learning models strongly relies on data quality, including carefully selected features based on established metrics such as Feature importance scores. This paper introduces PermuGini-RF, a hybrid feature selection and classification model that systematically combines Gini Importance for fast upstream screening and Permutation Importance for robust downstream validation...
Go to contribution page -
Rhea Senthil Kumar (University of California, San Diego)Poster
Gravitational-wave (GW) observations provide a unique probe of the underlying massive-star population, but extracting this information requires modeling how massive stars evolve into merging compact binaries across cosmic time. Forward population-synthesis frameworks that couple binary evolution to cosmic star-formation and metallicity histories are therefore needed to connect observed merger...
Go to contribution page -
Adnan EghtesadPoster
We introduce a physics-informed elasto-viscoplastic (NN-EVP) framework that utilizes Input Convex Neural Networks (ICNNs) to ensure thermodynamic consistency while maintaining high predictive expressivity. Developed within the PyTorch ecosystem, this automated constitutive modeling tool is validated against both synthetic power-law data and experimental uniaxial deformation data under large...
Go to contribution page -
Gabriele TrottaPoster
As artificial intelligence becomes more capable, it becomes ever more widely adopted, and so do their energy demands. Custom hardware such as FPGAs offers a way to absorb this cost, unveiling a rich design space across which a neural network can be tuned for competing objectives like accuracy, trustworthiness and power. Exploring that space comes with a caveat, however: the processing needed...
Go to contribution page -
Sijan Shrestha (Howard University)Poster
LUT-based Neural Networks (NNs) demonstrate significant potential for low-latency and high-throughput inference on FPGAs in the fields like high-energy physics, high-frequency trading, etc [1][2]. Ensemble approaches like AmigoLUT [3] improve scalability and accuracy of LUT-based NNs such as LogicNets [4] and NeuraLUT [5]. However, increasing ensemble size, even when we start with a small base...
Go to contribution page -
Timothy Sherwood (UC Santa Barbara)Presentation
As quantized neural network inference moves from cloud to edge, the demand for custom hardware accelerators tailored to specific models, precisions, and deployment constraints continues to grow. While frameworks for training and deploying quantized models in software are mature, the path from a trained model to working custom hardware remains steep and requires expertise spanning machine...
Go to contribution page -
Ms Akbota Assan (University of California, San Diego)Presentation
The search for axion dark matter requires recovering extremely weak, narrowband signals buried in noise, often under stringent latency and throughput constraints at the data-acquisition stage. We present a WaveNet-based architecture for denoising and detecting axion-like signals on the TIDMAD benchmark, derived from the MIT ABRACADABRA experiment. The model uses dilated causal convolutions and...
Go to contribution page -
Jose Daniel Gaytan Villarreal (Carnegie-Mellon University (US))Presentation
Ionizing radiation from cosmic rays and ambient radioactivity induces charge jumps in superconducting qubits, causing correlated errors that challenge fault-tolerant quantum computing while simultaneously providing a detection signature for quantum sensing applications. Current detection methods operate offline, providing only a post-hoc record of charge events. By the time a jump is...
Go to contribution page -
Leo RozanovPresentation
Zero-shot learning (ZSL) refers to the ability of a model to be able to classify unseen labels. We present High Granularity Quantization (HGQ) Linformer based event-level ZSL architecture model trained using contrastive methods on the Collide-2V dataset. The output of the model is an embedding vector which enables classification through ZSL. Different physics processes occupy unique regions in...
Go to contribution page -
Mr Kaamesh Chandrasekaran (Sri Venkateswara College of Engineering), Ms Madhushree Naga (Sri Venkateswara College of Engineering)Poster
Multi-messenger astronomy relies on three independent alert systems - gravitational wave detectors (LIGO/Virgo/KAGRA), high-energy neutrino observatories (IceCube), and gamma-ray monitors (Fermi-GBM), each firing alerts only when their own threshold is individually crossed. The problem is that when all three show near-threshold activity around the same time, no existing pipeline recognizes it...
Go to contribution page -
Aiken Xie (Columbia University), Jose Daniel Gaytan Villarreal (Carnegie-Mellon University (US))Presentation
We evaluate the FPGA deployment of a quantized recurrent probabilistic neural network (RPNN) online state estimator for real-time tokamak control. To navigate the strict resource and latency limitations of fusion environments, we detail a workflow utilizing quantization-aware training (QAT) via the Brevitas framework, followed by translation into an efficient FPGA implementation using a...
Go to contribution page -
Sagar Addepalli (SLAC National Accelerator Laboratory (US))Presentation
Real-time Quantum Inspired Algorithms for Anomaly Detection in Collider Triggers
Go to contribution page
Anomaly detection algorithms deployed at the edge of particle collider experiments offer a model-agnostic approach to search for signs of new physics, complementing typical topology- or energy-driven selections. Tensor networks are a class of quantum-inspired machine learning models that represent information in... -
Leonid DidukhPoster
Future high-energy physics and gravitational-wave experiments are projected to generate data at unprecedented event rates, demanding fast, scalable, and efficient data management systems. Because scientific data volumes continue to outpace available storage infrastructure, experimental workflows increasingly rely on a combination of real-time triggering mechanisms to filter uninformative...
Go to contribution page -
Jun-Sik YooPoster
Low-bit quantization is usually judged by whether the compressed model still works: perplexity, output KL, downstream accuracy, or layer reconstruction error. These are useful metrics, but they are also coarse views of a high-dimensional computation. A model can look healthy under these summaries while some part of its internal layer update has changed in a more structured way.
We study...
Go to contribution page -
75. Scaling Out hls4ml for ASICs: A Surrogate Model for Neural Network Accelerator Synthesis MetricsArghya Ranjan Das (Purdue University)Presentation
Custom ASIC accelerators offer significant power and performance advantages for machine learning in scientific and edge computing; a driving example is superconducting qubit readout, where moving real-time classification of qubit states from room-temperature FPGAs into the cryostat requires custom ASICs on cryo-compatible technology nodes. However, obtaining accurate area and timing requires...
Go to contribution page -
Prof. Gert Cauwenberghs (UC San Diego)
-
Andy MezaPresentation
When deployed in edge applications, neural networks (NNs) undergo numerous changes to ensure they adhere to strict power, performance, and size constraints while simultaneously being robust to faults. In prior work, NN robustness is evaluated using a bit-level ranking based on how sensitive an edge NN is to a fault in a given parameter bit. Unfortunately, the fault injection (FI) campaigns...
Go to contribution page -
Daniel Abadjiev (University of Chicago (US))Presentation
At a 10TeV muon collider, a high level of beam induced background (BIB) will overlay signal produced from muon collisions, similar to how pile-up at the HL-LHC will overlay signal from proton-proton collisions. On-detector differentiation of BIB from signal would improve performance of the inner pixel tracker. Following the successful development of a prototype “smartpixel” application...
Go to contribution page -
Jason Weitz (Univ. of California San Diego (US))Poster
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...
Go to contribution page -
Yuan-Tang Chou (University of Washington (US))Topical session
We would like to invite participants interested in working on or exploring the use of Inference as a Service in their experiments to present their use cases and ideas. The goal is to develop a Fast ML community vision that supports and encourages cross-experiment R&D efforts. The session is expected to consist of a series of lightning talks, with plenty of discussion in between to foster new...
Go to contribution page -
Mr Eric Jin (University of California, Berkeley)Presentation
Real-time decoding is the binding constraint for fault-tolerant quantum computing: superconducting qubits emit a syndrome round roughly every microsecond, and a decoder that cannot keep pace incurs an exponentially growing backlog. Neural decoders have reached an accuracy regime that classical methods (Union-Find, MWPM, belief propagation) miss, but few have been demonstrated within the...
Go to contribution page -
Ho-Fung Tsoi (University of Pennsylvania)Poster
Modern particle physics experiments often impose strict latency constraints (microseconds or below) on the edge electronics to extract quality signals from noisy raw data in real time. For 2D image data where signals are spatially sparse, standard CNNs are inefficient because latency and resources scale directly with image size, as every input pixel is densely convolved, including the vast...
Go to contribution page -
Ho-Fung Tsoi (University of Pennsylvania)Presentation
In collider experiments, particle identification (PID) in drift chambers has traditionally relied on the ionization energy loss ($dE/dx$), whose resolution is fundamentally limited by large Landau fluctuations. Instead, cluster counting ($dN/dx$) measures the number of primary ionization clusters, which is Poisson in nature with smaller statistical fluctuations and can offer higher separation...
Go to contribution page -
Angelos Ioannou (Lawrence Berkeley National Laboratory), Giuseppe Di Guglielmo (Fermi National Accelerator Laboratory), John Shalf (Lawrence Berkeley National Laboratory)Presentation
Developing efficient hardware accelerators for mathematical kernels remains a labor-intensive process, traditionally requiring deep expertise in low-level hardware description languages and extensive manual optimization. While high-level synthesis tools and domain-specific hardware generators have improved productivity, the space of achievable designs for any given kernel is vast, and...
Go to contribution page -
Aarav GaurPoster
On-probe spike sorting aims to discriminate between neurons using electrical signals recorded directly at the probe, enabling real-time applications such as brain-computer interfaces. To generate ground-truth training data, we follow SpikeForest/MEArec-style simulation methodology to construct a synthetic multi-neuron tetrode recording with physiologically-motivated waveforms, controlled SNR,...
Go to contribution page -
Akbota Assan (University of California, San Diego)Poster
Axion dark matter searches such as ABRACADABRA produce continuous high-
Go to contribution page
rate time series in which injected signals occupy a single narrow frequency bin
per time frame, a structure that is natural in the spectral domain but opaque in
the raw time domain. Existing TIDMAD denoising approaches either operate
directly on raw time series or require separate model weights per frequency... -
Dmitry Kondratyev (Purdue University (US))Presentation
The increasing ubiquity and complexity of machine learning models in scientific workflows motivate the adoption of inference-as-a-service architectures, which enable efficient utilization of hardware accelerators. The SONIC (Services for Optimized Network Inference on Coprocessors) framework provides a blueprint for implementing inference as a service in high-energy physics (HEP),...
Go to contribution page -
Maira Khan (Fermi National Accelerator Laboratory)Presentation
High-temperature superconducting (HTS) magnets are a key enabling technology for future muon colliders, fusion energy systems, and other high-field scientific facilities. Ensuring safe operation requires rapid estimation of thermal stability margins and thermal runaway thresholds, yet these quantities are typically obtained through iterative numerical solutions of nonlinear heat-balance...
Go to contribution page -
Pritam Palit (Carnegie-Mellon University (US))Poster
DeepTau is the convolutional neural network (CNN)-based multiclass classifier for hadronic tau identification in CMS. To improve inference performance and simplify deployment, the DeepTau models have been migrated from TensorFlow to ONNX (Open Neural Network Exchange) within the CMS software framework (CMSSW) using ONNX Runtime. In parallel, the TensorFlow-based deployment in SONIC (Services...
Go to contribution page -
Dr Antonio Vagnerini (University of Nebraska-Lincoln)Poster
The increasing complexity and data throughput of the CMS experiment at the LHC demand scalable and intelligent tools to ensure data quality. In this talk, we present a machine learning-oriented infrastructure designed to support the offline data quality monitoring (DQM) process at CMS. The infrastructure enables the integration of ML algorithms into the DQM workflow, providing auto- mated...
Go to contribution page -
Sterre Hoogendoorn (University of Pennsylvania)Poster
Anomaly detection (AD) has recently emerged as an exciting alternative to conventional search strategies in high energy physics. The integration of these techniques into trigger systems is even more recent, but represents a crucial step in expanding the coverage of LHC triggers. In this paper, we explore the direct comparison, as well as combination, of two compression techniques for...
Go to contribution page -
Seokju Chung (Columbia University)Presentation
Raw wire readout data from Liquid Argon Time Projection Chambers (LArTPCs) in future experiments such as the Deep Underground Neutrino Experiment (DUNE) will vastly exceed available storage bandwidth, necessitating intelligent online data selection. We present a compressed autoencoder architecture for model-independent anomaly detection operating directly on LArTPC wire waveforms, benchmarked...
Go to contribution page -
Adrian Chitan (IFIN-HH (RO))Presentation
High-energy physics experiments face extreme data rates, requiring real-time trigger systems to reduce event throughput while preserving sensitivity to rare processes. Trigger systems have traditionally been constructed as modular chains of sequentially optimised algorithms, each optimised for a specific local objective with no guarantee of overall optimality.
We instead formulate trigger...
Go to contribution page -
Marius Köppel (ETH Zurich (CH))Poster
The CMS Phase-2 upgrade integrates machine learning (ML) throughout the Level-1 Trigger, marking a transition toward differentiable detector systems in which multiple ML models collaboratively reconstruct physics objects in real time. As detector conditions and physics goals evolve, these models require continuous retraining, validation, and deployment, transforming trigger algorithms into...
Go to contribution page -
Subhashini Sivagnanam (San Diego Supercomputer Center), Yuwu Chen (San Diego Supercomputer Center)Poster
The Triton Shared Computing Cluster (TSCC) at the San Diego Supercomputer Center is evolving beyond a traditional high-performance computing system into an AI-for-science platform that enables researchers to integrate artificial intelligence into scientific discovery. TSCC provides campus researchers with access to computational resources through both Condo (system purchase) and Hotel...
Go to contribution page -
Chang Sun (California Institute of Technology (US))Tutorial
Real-time inference with sub-microsecond latency is critical for the Level-1 trigger systems at the High-Luminosity LHC. We present an end-to-end, open-source framework that spans model optimization, quantization, and FPGA deployment, enabling the translation of high-level neural network or generic dataflow models into resource-efficient FPGA implementations.
We intro HGQ and Alkaid, the...
Go to contribution page -
Philip Coleman Harris (Massachusetts Inst. of Technology (US))Presentation
Ultrafast online learning is essential for high-frequency systems, such as controls for quantum computing and nuclear fusion, where adaptation must occur on sub-microsecond timescales. Meeting these requirements demands low-latency, fixed-precision computation under strict memory constraints, a regime in which conventional networks buil on Multi-Layer Perceptrons (MLPs) are both inefficient...
Go to contribution page -
Julia HaynesPoster
The growing volume of substellar spectra from JWST, including NIRSpec observations of brown dwarfs and directly imaged exoplanets, demands increasingly efficient tools for atmospheric characterization. Traditional spectral fitting approaches such as grid interpolation and Markov Chain Monte Carlo (MCMC) retrieval become significant computational bottlenecks when applied to large samples or...
Go to contribution page -
Arianna Cox (Imperial College (GB))Presentation
Transformers are promising for real-time intelligent systems, but their arithmetic complexity makes microsecond-scale FPGA inference challenging. This paper presents a LUT-based Transformer framework that combines heterogeneous quantization and LUT-aware training to map compact Transformer models into hardware-efficient lookup-table structures. The proposed method jointly optimizes precision,...
Go to contribution page
Choose timezone
Your profile timezone: