Engineering Papers⌕ Search

SEARCH · Engineering Papers

Results for “machine learning for scientific computing”

Search indexed NASA NTRS and DOE OSTI research on propulsion, heat transfer, battery materials and energy systems. Follow report and document links to the original sources.

Quote a phrase for an exact phrase match. Source license links do not imply unrestricted reuse.

At least 487 records · Page 27
All sourcesNASA NTRSDOE OSTI
DOE OSTI2025

Exocortex Network for AI-Augmented Human-Led Scientific Expedition

AI advances in science can be viewed along two main directions with a fluid boundary: enhancing efficiency through automation and smart tools to accelerate tasks that humans can already perform; and enabling exploration into uncharted territories and potentially toward AGI. These advances manifest in the AI cognitive core through the development and explainability of foundation models; in the physical embodiment of instruments and facilities; and in the integrated agency of AI workflows exemplified by the science exocortex. To address the role of humans in this evolving landscape, in this Perspective, we suggest a third direction: the development of personalized agents that form human-centered networks, supporting both efficiency and exploration while ensuring that AI remains aligned with human vision.

97 MATHEMATICS AND COMPUTING↗
DOE OSTI2024

Deep Generative Models in Energy System Applications: Review, Challenges, and Future Directions

In recent years, with the advent of mature machine learning products like ChatGPT, Stable Diffusion, and Sora, the world has witnessed tremendous changes driven by the rapid development of generative artificial intelligence (GAI). Beyond applications in text, speech, image, and video creation, deep generative models (DGMs) underpinning these cutting-edge technologies have also been employed by domain researchers to address scientific and engineering challenges. This paper aims to fill a gap in the research community by providing a systematic review of how DGMs have been utilized in energy system applications. After introducing four most popular DGMs, we review and categorize 196 research articles into five focus areas: data generation, forecasting, situational awareness, modeling, and optimal decision-making. Through this classification, we uncover trends in how DGMs are employed for each type of problem, highlighting GAI techniques that contribute to breakthroughs over traditional methods. We discuss limitations in existing literature, engineering challenges, and propose future directions, all tailored to the unique nature of problems in energy system engineering. Our goal is to offer insights for energy system domain researchers, providing a comprehensive view of existing studies and potential future opportunities.

32 ENERGY CONSERVATION, CONSUMPTION, AND UTILIZATI↗
DOE OSTI2026

HydraGNN_Predictive_GFM_2026 - Ensemble of predictive graph foundation models for atomistic materials modeling

This release contains data and parameters of HydraGNN-based graph foundation models trained as a result of the work published in the pre-print "Exascale Multi-Task Graph Foundation Models for Imbalanced, Multi-Fidelity Atomistic Data" by M. Lupo Pasini et al. (https://arxiv.org/abs/2604.15380). We jointly train on 16 open first-principles datasets (544+ million structures covering 85+ elements) using a multi-task architecture with per-dataset heads and a scalable ADIOS2/DDStore data pipeline. On Frontier, we execute six large-scale DeepHyper hyperparameter optimization campaigns in FP64 and promote the top-performing message-passing models to sustained 2,048-node training, yielding a PaiNN-based lead model. The version of HydraGNN used to generate the outputs provided in this release is HydraGNN v5.0 (https://github.com/ORNL/HydraGNN/releases/tag/v5.0) The list of datasets used for the training of the graph foundation model is the following: 1) Alexandria [1] 2) ANI1x [2] 3) MPTrj [3] 4) Open Catalyst 2020 (OC20) [4] 5) Open Catalyst 2022 (OC22) [5] 6) Open Catalyst 2025 (OC25) [6] 7) Open Direct ir Capture 2023 (ODAC23) [7] 8) Open Materials 2024 (OMat24) [8] 9) Open Molecules 2025 (OMol25) [9] 10) OMol25-neutral (subset of OMol25 that contains only molecules with zero total charge) 11) OMol25-non-neutral (subset of OMol25 that contains only molecules with non-zero total charge) 12) Open Polymers 2026 (OPoly2026) [10] 13) Nabla2DFT [11] 14) QCML [12] 15) QM7X [reference 13] 16) transition1x [14] Dataset references: [1] J. Schmidt et al., “A dataset of 175k stable and metastable materials calculated with the PBEsol and SCAN functionals,” Scientific Data, vol. 9, p. 64, 2022. [2] J. S. Smith et al., “The ANI-1ccx and ANI-1x data sets, coupled-cluster and density functional theory properties for molecules,” Scientific Data, vol. 7, p. 134, 2020. [Online]. Available: https: //www.nature.com/articles/s41597-020-0473-z [3] A. Jain et al., “Commentary: The Materials Project: A materials genome approach to accelerating materials innovation,” APL Materials, vol. 1, no. 1, p. 011002, 07 2013. [Online]. Available: https://doi.org/10.1063/1.4812323 [4] L. Chanussot et al., “Open catalyst 2020 (oc20) dataset and community challenges,” ACS Catalysis, vol. 11, no. 10, pp. 6059–6072, 2021. [Online]. Available: https://doi.org/10.1021/acscatal.0c04525 [5] K. Tran et al., “Open catalyst 2022 (oc22) dataset and challenges for oxidation electrocatalysts,” ACS Catalysis, vol. 13, no. 5, pp. 3066–3084, 2023. [Online]. Available: https://doi.org/10.1021/acscatal.2c05426 [6] S. J. Sahoo et al., “The open catalyst 2025 (oc25) dataset and models for solid-liquid interfaces,” arXiv preprint arXiv:2509.17862, 2025. [Online]. Available: https://arxiv.org/abs/2509.17862 [7] A. Sriram et al., “The open DAC 2023 dataset and challenges for sorbent discovery in direct air capture,” ACS Central Science, vol. 10, no. 5, pp. 923–941, 2024. [8] L. Barroso-Luque et al., “Open materials 2024 (omat24) inorganic materials dataset and models,” 2024. [Online]. Available: https://arxiv.org/abs/2410.12771 [9] D. S. Levine et al., “The open molecules 2025 (OMol25) dataset, evaluations, and models,” 2025. [Online]. Available: https://arxiv.org/abs/2505.08762 [10] D. S. Levine et al., The open polymers 2026 (OPoly26) dataset and evaluations,” arXiv preprint arXiv:2512.23117, 2025. [Online]. Available: https://arxiv.org/abs/2512.23117 [11] K. Khrabrov et al., “Nabla2dft: A universal quantum chemistry dataset of drug-like molecules and a benchmark for neural network potentials,” in NeurIPS 2024 Datasets and Benchmarks Track, 2024. [Online]. Available: https://openreview.net/forum?id=ElUrNM9U8c [12] S. Ganscha et al., “The QCML dataset, quantum chemistry reference data from 33.5M DFT and 14.7B semi-empirical calculations,” Scientific Data, vol. 12, p. 406, 2025. [13] J. Hoja et al., “QM7-X, a comprehensive dataset of quantum-mechanical properties spanning the chemical space of small organic molecules,” Scientific Data, vol. 8, p. 43, 2021. [Online]. Available: https://www.nature.com/articles/s41597-021-00812-2 [14] M. Schreiner et al., “Transition1x - a dataset for building generalizable reactive machine learning potentials,” Scientific Data, vol. 9, p. 779, 2022. The folder "datasets_ADIOS2_format" contains the set of pre-processed datasets in Adaptable I/O System (ADIOS) format (https://www.exascaleproject.org/research-project/adios/) that have been used for the development and training of GFMs in this work. The "datasets_ADIOS2_format" directory contains 2 sub-directories, one for the version "v1" of the datasets and one for the version "v2" of the datasets. The version "v1" of the datasets provides values of the total energy as they are extracted from the original data as it was released by the respective institutions. The version "v2" of the datasets provides values of the energy that have been realigned. The realignment was performed by training a linear regression model that predicts the total energy as a function of the chemical composition of the atomistic structure, and then subtract such prediction from the original value of the total energy. Both folders "v1" and "v2" contain 16 sub-directories, each corresponding to an ADIOS2-formatted dataset The folder "DeepHyper-results" contains the configurational files and model's parameters for all the 186 HPO trials that were successfully completed by the scalable hyperparameter optimization (HPO) runs on Frontier. The content of the folder "DeepHyper-results" I structured as follows: 1) task-list.txt: list of mpnn name, jobid, and deephyper task id 2) gfm_${MPNN}_${JOBID}_0.${TASKID}: run directory with checkpoint files 3) gfm_${MPNN}: deephyper summary directory (*.csv) for each specific MPNN type 4) deephyper-experiment-${JOBID}: output and error logs for each job The file "deephyper-sorted.csv" contains the details of each HydraGNN model built and tested by HPO, obtained by merging the (*.csv) filed from each HPO run executed. Out of all the HPO trials, we selected 10 to continue the training of the respective HydraGNN models. Due to limited computational budget available in the LRN070 allocation we could not complete the training till convergence for all these 10 selected models. The folder "models" contains multiple sub-folders, one per each HydraGNN model trained. Each model sub-folder contains the parameters of each HydraGNN model, with multiple checkpoint-restarts. The list of sub-folders are as follows: 1) multidataset_hpo-BEST1-fp64 2) multidataset_hpo-BEST2-fp64 3) multidataset_hpo-BEST3-fp64 4) multidataset_hpo-BEST4-fp64 5) multidataset_hpo-BEST5-fp64 6) multidataset_hpo-BEST6-fp64 7) multidataset_hpo-BEST7-fp64 8) multidataset_hpo-BEST8-fp64 9) multidataset_hpo-BEST9-fp64 10) multidataset_hpo-BEST10-fp64 Within each one of these folders, additional auxiliary log files are provided with descriptions about how the training proceeded. The lead PaiNN-model is contained inside "multidataset_hpo-BEST6-fp64". The file "mlp_branch_weights" contains the parameters of the multi-layer perceptron (MLP) used to reconcile the predictions of the 16 output decoding heads of the HydragNN architectures. The MLP takes in input the chemical composition of the atomistic structure and predicts averaging weights to linearly mix the predictions of each output decoding head toward consolidating them into a single one. The folder "1.1billion-structure-inference" contains 1.1 billion atomistic structures randomly generated. Each structures is associated with energy and forces predicted with the lead-PaiNN model combined with the MLP model for reconciliation of the multi-branch predictions generated by the 16 output decoding heads. The folder "1.1billion-structure-inference" contains 9,300 (*.tar.gz) subdirectories, one per Frontier compute node used to execute the inference at exascale. Once uncompressed, each (*.tar.gz) subdirectory contains an ADIOS2 (*.bp) file container, where each atomistic structure is stored as a PyTorch-Geometric Data object. The file "export_dataset_environment_variables.sh" contains the environment variables that need to be set before running the HydraGNN code to reproduce the results provided in this dataset release. The code that can be used to load the ADIOS2 files, load HydraGNN models, and run inference is available at: https://github.com/ORNL/HydraGNN/releases/tag/v5.0

36 MATERIALS SCIENCE↗
DOE OSTI2024

Coincident learning for unsupervised anomaly detection of scientific instruments

Abstract Anomaly detection is an important task for complex scientific experiments and other complex systems (e.g. industrial facilities, manufacturing), where failures in a sub-system can lead to lost data, poor performance, or even damage to components. While scientific facilities generate a wealth of data, labeled anomalies may be rare (or even nonexistent), and expensive to acquire. Unsupervised approaches are therefore common and typically search for anomalies either by distance or density of examples in the input feature space (or some associated low-dimensional representation). This paper presents a novel approach called coincident learning for anomaly detection (CoAD), which is specifically designed for multi-modal tasks and identifies anomalies based on coincident behavior across two different slices of the feature space. We define an unsupervised metric, F ^ β , out of analogy to the supervised classification F β statistic. CoAD uses F ^ β to train an anomaly detection algorithm on unlabeled data , based on the expectation that anomalous behavior in one feature slice is coincident with anomalous behavior in the other. The method is illustrated using a synthetic outlier data set and a MNIST-based image data set, and is compared to prior state-of-the-art on two real-world tasks: a metal milling data set and our motivating task of identifying RF station anomalies in a particle accelerator.

43 PARTICLE ACCELERATORS↗
DOE OSTI2024

Physics-informed neural networks for heterogeneous poroelastic media

This study presents a novel physics-informed neural network (PINN) framework for modeling poroelasticity in heterogeneous media with material interfaces. The approach introduces a composite neural network (CoNN) where separate neural networks predict displacement and pressure variables for each material. While sharing identical activation functions, these networks are independently trained for all other parameters. To address challenges posed by heterogeneous material interfaces, the CoNN is integrated with the Interface-PINNs (I-PINNs) framework (Sarma et al., Comput. Methods Appl. Mech. Eng. 429: 117135, 2024), allowing different activation functions across material interfaces. Further, this ensures accurate approximation of discontinuous solution fields and gradients. Performance and accuracy of this combined architecture were evaluated against the conventional PINNs approach, a single neural network (SNN) architecture, and the eXtended PINNs (XPINNs) framework through two one-dimensional benchmark examples with discontinuous material properties. The results show that the proposed CoNN with I-PINNs architecture achieves an RMSE that is two orders of magnitude better than the conventional PINNs approach and is at least 40 times faster than the SNN framework. Compared to XPINNs, the proposed method achieves an RMSE at least one order of magnitude better and is 40% faster.

42 ENGINEERING↗
DOE OSTI2025

Leveraging structure-informed machine learning for fast steric zipper propensity prediction across whole proteomes

Predicting the amyloid fold and the propensity of peptide segments to adopt amyloid-like structures remain a challenge. However, recent progress has facilitated structure-based prediction of steric zipper propensity and the use of machine learning to accelerate the calculation of predictive models across many scientific areas. Leveraging these advances, we have developed a new approach for rapid proteome-wide assessment of zipper profiles that is informed by four million steric zipper predictions collected over ten years. This collection is used to build a machine learning model capable of rapidly predicting steric zipper propensity, and allowing for the assessment of zippers at both the protein and proteome level. Our predictions show enrichment for zipper forming segments in proteins involved in cell wall reorganization in yeast, highlighting a potential category of interest for experimental characterization. Overall, our predictive model allows for the exploration of amyloid formation across the tree of life and provides a tool for assessment of both novel and designed sequences for zipper density.

Biochemistry & Molecular Biology↗
DOE OSTI2021

Train Like a (Var)Pro: Efficient Training of Neural Networks with Variable Projection

Deep neural networks (DNNs) have achieved state-of-the-art performance across a variety of traditional machine learning tasks, e.g., speech recognition, image classification, and segmentation. The ability of DNNs to efficiently approximate high-dimensional functions has also motivated their use in scientific applications, e.g., to solve partial differential equations and to generate surrogate models. In this paper, we consider the supervised training of DNNs, which arises in many of the above applications. We focus on the central problem of optimizing the weights of the given DNN such that it accurately approximates the relation between observed input and target data. Devising effective solvers for this optimization problem is notoriously challenging due to the large number of weights, nonconvexity, data sparsity, and nontrivial choice of hyperparameters. To solve the optimization problem more efficiently, we propose the use of variable projection (VarPro), a method originally designed for separable nonlinear least-squares problems. Our main contribution is the Gauss--Newton VarPro method (GNvpro) that extends the reach of the VarPro idea to nonquadratic objective functions, most notably cross-entropy loss functions arising in classification. These extensions make GNvpro applicable to all training problems that involve a DNN whose last layer is an affine mapping, which is common in many state-of-the-art architectures. In our four numerical experiments from surrogate modeling, segmentation, and classification, GNvpro solves the optimization problem more efficiently than commonly used stochastic gradient descent (SGD) schemes. Finally, GNvpro finds solutions that generalize well, and in all but one example better than well-tuned SGD methods, to unseen data points.

97 MATHEMATICS AND COMPUTING↗
DOE OSTI2023

Trust-Enhancing Probabilistic Transfer Learning for Sparse and Noisy Data Environments

There is an increasing aspiration to utilize machine learning (ML) for various tasks of relevance to national security. ML models have thus far been mostly applied to tasks and domains that, while impactful, have sufficient volume of data. For predictive tasks of national security relevance, ML models of great capacity (ability to approximate nonlinear trends in input-output maps) are often needed to capture the complex underlying physics. However, scientific problems of relevance to national security are often accompanied by various sources of sparse and/or incomplete data, including experiments and simulations, across different regimes of operation, of varying degrees of fidelity, and include noise with different characteristics and/or intensity. State-of-the-art ML models, despite exhibiting superior performance on the task and domain they were trained on, may suffer detrimental loss in performance in such sparse data environments. This report summarizes the results of the Laboratory Directed Research and Development project entitled Trust-Enhancing Probabilistic Transfer Learning for Sparse and Noisy Data Environments. The objective of the project was to develop a new transfer learning (TL) framework that aims to adaptively blend the data across different sources in tackling one task of interest, resulting in enhanced trustworthiness of ML models for mission- and safety-critical systems. The proposed framework determines when it is worth applying TL and how much knowledge is to be transferred, despite uncontrollable uncertainties. The framework accomplishes this by leveraging concepts and techniques from the fields of Bayesian inverse modeling and uncertainty quantification, relying on strong mathematical foundations of probability and measure theories to devise new uncertainty-aware TL workflows.

97 MATHEMATICS AND COMPUTING↗
DOE OSTI2024

Machine Learning for Predictive Performance Analysis in Charged Particle Beam Tools

Imaging methods driven by probes, electrons, and ions have played a dominant role in modern science and engineering. Opportunities for machine vision and AI that focus on consumer problems like driving and feature recognition, are now presenting themselves for automating aspects of the scientific processes. This proposal aims to enable and drive discovery in ultra-low energy implantation by taking advantage of faster processing, flexible control and detection methods, and architecture-agnostic workflows that will result in higher efficiency and shorter scientific development cycles. Custom microscope control, collection and analysis hardware will provide a framework for conducting novel in situ experiments revealing unprecedented insight into surface dynamics at the nanoscale. Ion implantation is a key capability for the semiconductor industry. As devices shrink, novel materials enter the manufacturing line, and quantum technologies transition to being more mainstream. Traditional implantation methods fall short in terms of energy, ion species, and positional precision. Here we demonstrate 1 keV focused ion beam Au implantation into Si and validate the results via atom probe tomography. We show the Au implant depth at 1 keV is 0.8 nm and that identical results for low energy ion implants can be achieved by either lowering the column voltage, or decelerating ions using bias – while maintaining a sub-micron beam focus. We compare our experimental results to static calculations using SRIM and dynamic calculations using binary collision approximation codes TRIDYN and IMSIL. A large discrepancy between the static and dynamic simulation is found that is due to lattice enrichment with high stopping power Au and surface sputtering. Additionally, we demonstrate how model details are particularly important to the simulation of these low-energy heavy-ion implantations. Finally, we discuss how our results pave a way to much lower implantation energies, while maintaining high spatial resolution.

47 OTHER INSTRUMENTATION↗
DOE OSTI2025

The Artificial Intelligence Ontology: LLM-Assisted Construction of AI Concept Hierarchies

The Artificial Intelligence Ontology (AIO) is a systematization of artificial intelligence (AI) concepts, methodologies, and their interrelations. Developed via manual curation, with the additional assistance of large language models (LLMs), AIO aims to address the rapidly evolving landscape of AI by providing a comprehensive framework that encompasses both technical and ethical aspects of AI technologies. The primary audience for AIO includes AI researchers, developers, and educators seeking standardized terminology and concepts within the AI domain. We use the term “branches” for classes, and their subclasses, in our ontology that are subclasses of owl:Thing. AIO contains eight branches: Bias, Layer, Machine Learning Task, Mathematical Function, Model, Network, Preprocessing, and Training Strategy, each designed to support the modular composition of AI methods and facilitate a deeper understanding of deep learning architectures and ethical considerations in AI. AIO uses the Ontology Development Kit (ODK) for its creation and maintenance, with its content being more easily updated through AI-driven curation support. This approach not only ensures the ontology's relevance amidst the fast-paced advancements in AI but also significantly enhances its utility for researchers, developers, and educators by simplifying the integration of new AI concepts and methodologies. The ontology's utility is demonstrated through the annotation of AI methods data in a catalog of AI research publications and the integration into the BioPortal ontology resource, highlighting its potential for cross-disciplinary research. The AIO ontology is open source and is available on GitHub ( https://w3id.org/aio/ ) and BioPortal ( https://bioportal.bioontology.org/ontologies/AIO ).

Joachimiak, Marcin P. [Biosystems Data Science Dep↗
DOE OSTI2021

SIRIUS: Science-Driven Data Management for Multi-Tiered Storage

The data sets being generated by large applications on very large-scale systems are increasing in both size and complexity. At the same time, there are new ways available to store and access these data sets. The goal in this project is to develop software that applications can use to make use of new and existing storage technologies in more sophisticated ways. One challenge in scientific data management is handling ‘hot’ vs ‘cold’ data. Data that is hot is data that is needed (or will be needed soon) in order for the program to continue progressing, while cold data is either output (and so will not be need further during the life of the program) or will not be needed until significantly later in the program’s run. Hot data should be stored in a way that allows fast access. On most systems, economic factors lead to an inverse relationship between storage performance and storage capacity and so fast access storage is limited. This makes it important to correctly place hot and cold data and avoid cold data unnecessarily consuming precious resources. In this reporting period, we addressed this challenge in various ways and at various levels. Data management frameworks offer only limited control to applications in how data is stored. We have added software capabilities for seamlessly moving data between layers of the storage technology using promote and demote functions to existing software frameworks. This gives direct control to applications in deciding what priority data receives. Additionally, we integrated different storage layer management frameworks in order to allow data to be exchanged and moved between storage layers in a consistent way across the application. Further, applications are not always able to directly decide what storage level makes sense for a given piece of data without an understanding of the underlying storage technologies. Data storage frameworks are often positioned to make these sorts of decisions in service of the application. We have added machine-learning based capabilities to data staging frameworks in order to make intelligent decisions about where data should be stored given learning about patterns in previous usage of similar data.

97 MATHEMATICS AND COMPUTING↗
DOE OSTI2020

Uncertainty Quantification for Capacity Expansion Planning

This report quantifies the uncertainty in output decisions from a Capacity Expansion Planning (CEP) model. The need to understand how uncertainties within CEP models and modeling assumptions affect Quantities of Interest (QoIs) such as expansion and operating costs, as well as expansion decisions remains an ongoing challenge in scientific research and industrial operations. This area of research is particularly important for models which seek to capture how large networks will evolve and operate under increased sources of variable generation, i.e., higher penetration of renewable technologies such as solar and wind generators. Uncertainty quantification (UQ) of CEP models which estimate expansion costs and decisions, and production cost models which estimate operating costs and dispatch decisions, is a key focus of research at NREL. The Regional Energy Deployment System (ReEDS) represents a state-of-the-art CEP model and considers a range of possible grid evolutions in an attempt to identify key drivers, ramifications, and decisions which contribute to better informed investment and policy decisions. However, research to quantify how uncertainties and model assumptions, such as unit commitment (UC), within ReEDS may be affecting its outputs remains challenging due to to size and complexity of the model

24 POWER TRANSMISSION AND DISTRIBUTION↗
DOE OSTI2020

High-throughput investigation of the formation of double spinels

Spinel compounds, with the general chemical formula AB 2 O 4 , are a wide class of materials, where A and B can be a variety of cations, providing this structure with a great deal of functional flexibility and giving rise to its considerable scientific interest. Recently, a spinel with the general formula ABB'O 4 has been predicted, increasing the possible usability of the spinels due to the higher cation diversity in the so-called double spinel structure. In this study, we use density functional theory calculations to predict if double spinels can be formed between experimentally synthesized normal and inverse single spinels. Our computations reveal that 49 double spinels have negative mixing enthalpies and are thus thermodynamically stable, with most of the stable compounds being formed from one of two distinct cation orderings. We show that the 17 different cations that form the different double spinels have a preferred site, tetrahedral or octahedral, except for Mn, Fe and Co which can occupy both sites interchangeably. We also study the relation between mixing enthalpies and cation-specific properties, as well as ways to classify the double spinels into distinct types and spinel groups depending on the cation ordering and cation oxidation states, respectively. Furthermore, by applying the Sure Independence Screening and Sparsifying Operator (SISSO) approach on the coordination-dependent ionic radii of the elemental constituents, we show that an interplay of local strain and electrostatic dominated terms can be used to separate the double spinels into distinct structural types depending on the cation order and their oxidation states.

37 INORGANIC, ORGANIC, PHYSICAL, AND ANALYTICAL CH↗
DOE OSTI2025

FM4NPP: A Scaling Foundation Model for Nuclear and Particle Physics

Large language models have revolutionized artificial intelligence by enabling large, generalizable models trained through self-supervision. This paradigm has inspired the development of scientific foundation models (FMs). However, applying this capability to experimental particle physics is challenging due to the sparse, spatially distributed nature of detector data, which differs dramatically from natural language. This work addresses if an FM for particle physics can scale and generalize across diverse tasks. We introduce a new dataset with more than 11 million particle collision events and a suite of downstream tasks and labeled data for evaluation. We propose a novel self-supervised training method for detector data and demonstrate its neural scalability with models that feature up to 188 million parameters. With frozen weights and task-specific adapters, this FM consistently outperforms baseline models across all downstream tasks. The performance also exhibits robust data-efficient adaptation. Further analysis reveals that the representations extracted by the FM are task-agnostic but can be specialized via a single linear mapping for different downstream tasks.

72 PHYSICS OF ELEMENTARY PARTICLES AND FIELDS↗
DOE OSTI2024

Data-driven closure modeling for hypersonic turbulent flows

The Reynolds-averaged Navier–Stokes (RANS) equations remain a workhorse technology for simulating compressible fluid flows of practical interest. Due to model-form errors, however, RANS models can yield erroneous predictions that preclude their use on mission-critical problems. This report summarizes work performed from FY22-FY24 focused on improving RANS models for hypersonic flows using data-driven modeling and scientific machine learning. In this work we: 1. Investigate the current capabilities of RANS models in Sandia’s parallel aerodynamics and re-entry code (SPARC) for hypersonic flows with a focus on shock boundary layer interactions (SBLIs), 2. Assess several established corrections that exist in the literature aimed at improving predictions for SBLIs, 3. Develop improved models for the Reynolds stress tensor using tensor-basis neural networks, 4. Develop a neural-network-based variable turbulent Prandtl number model to reduce errors in wall heating in SBLIs. 5. Begin future investigations including employing the LIFE framework to improve wall heating predictions in SBLIs as well as the ensemble Kalman filter. We find that current RANS models in SPARC are deficient for complex SBLI flows. In particular, no current model jointly predicts wall heat flux, wall shear stress, and wall pressure with reasonable accuracy. Existing corrections help, but do not alleviate this issue altogether. The development of improved models for the Reynolds stress tensor via tensor-basis neural networks results in more predictive RANS models across a suite of low-speed and high-speed cases. For hypersonic boundary layers, the inclusion of the wall-normal Reynolds stress via TBNNs has an appreciable impact on the wall-normal momentum balance and wall quantities. However, we find that improvements to the Reynolds stress tensor do not address the over-prediction in wall heat flux in SBLIs. We find that a neural-network-based variable turbulent Prandtl number model systematically and substantially improves wall heating predictions for a range of SBLI cases.

97 MATHEMATICS AND COMPUTING↗
DOE OSTI2023

Beyond the Hype: An Evaluation of Commercially Available Machine-Learning-Based Malware Detectors

There is a lack of scientific testing of commercially available malware detectors, especially those that boast accurate classification of never-before-seen (i.e., zero-day) files using machine learning (ML). Consequently, efficacy of malware detectors is opaque, inhibiting end users from making informed decisions and researchers from targeting gaps in current detectors. In this paper, we present a scientific evaluation of four prominent commercial malware detection tools to assist an organization with two primary questions: To what extent do ML-based tools accurately classify previously and never-before-seen files? Is purchasing a network-level malware detector worth the cost? To investigate, we tested each tool against 3,536 total files (2,554 or 72% malicious, 982 or 28% benign) of a variety of file types, including hundreds of malicious zero-days, polyglots, and APT-style files, delivered on multiple protocols. We present statistical results on detection time and accuracy, consider complementary analysis (using multiple tools together), and provide two novel applications of the recent cost-benefit evaluation procedure of Iannacone & Bridges. Although the ML-based tools are more effective at detecting zero-day files and executables, the signature-based tool might still be an overall better option. Both network-based tools provide substantial (simulated) savings when paired with either host tool, yet both show poor detection rates on protocols other than HTTP or SMTP. Our results show that all four tools have near-perfect precision but alarmingly low recall, especially on file types other than executables and office files—37% of malware, including all polyglot files, were undetected. Priorities for researchers and takeaways for end users are given. Code for future use of the cost model is provided.

97 MATHEMATICS AND COMPUTING↗
DOE OSTI2025

Even Higher-Level Synthesis: An Exploration of AI Hardware Accelerators using HLS4ML

With the rise of artificial intelligence, the popularization of deep learning, and a constantly evolving industry, the demand for flexible and efficient tools has never been greater. As algorithms grow more complex, their runtime and energy consumption increase exponentially. Customized hardware accelerators, long used for specific mathematical operations, remain essential for managing modern applications' computational and power demands. Hardware accelerators can speed up complex computations by orders of magnitude, but their manual design and verification processes are often challenging and time-consuming. High-Level Synthesis (HLS) provides a solution by transforming high-level algorithm descriptions, typically written in C++ or SystemC, into synthesizable RTL suitable for hardware implementation. This approach reduces development time for RTL engineers while offering flexibility beyond what traditional handwritten RTL can provide. We extended this capability to the machine-learning domain with the open-source framework hls4ml, which allows neural networks trained in Python frameworks like Tensorflow or PyTorch to be synthesized into efficient hardware representations for the traditional FPGA and ASIC flows. This breakthrough addresses the growing need for reduced design turnaround and easy verification of ML hardware accelerators with low latency and power efficiency constraints. During this tutorial, we will demonstrate how Python complements HLS by simplifying the ML design process, bridging the gap between software and hardware development. Attendees will explore how we translate neural networks modeled in Python into fixed-point C++ models suitable for HLS workflows. We will dive into strategies like Value-Range Analysis and Quantization-Aware Training, which optimize these designs for deployment and evaluate their accuracy, power consumption, and energy efficiency. To exemplify these concepts, experts from Fermilab will share their experiences applying this technology to high-energy physics experiments, where real-time, low-latency processing is critical. Over the years, Fermilab engineers have demonstrated how deep neural networks, optimized for hardware using hls4ml, can meet the stringent requirements of trigger systems at the CERN Large Hadron Collider. These systems rely on rapid decision-making to process immense data volumes while retaining only the most relevant events for further analysis. The application of hls4ml has also been extended to innovative technologies like smart pixel arrays. These smart pixels integrate ML inference capabilities directly into sensor devices, enabling localized data processing at the pixel level. This approach drastically reduces the need to transmit raw data to external processing units, significantly decreasing power consumption and latency. By embedding neural networks within the pixel architecture, the smart pixels can identify and prioritize relevant data in real time, providing a highly efficient solution for edge computing in scenarios such as particle detectors and imaging systems. Fermilab's work highlights the potential of hardware-accelerated ML in scenarios where both speed and power efficiency are mission-critical. Through this tutorial, attendees will gain valuable insights into the challenges and solutions of deploying ML in hardware. Understanding how HLS and hls4ml streamline the development of neural network-based hardware accelerators is fundamental for the industry's future. Participants will learn how these technologies are shaping the future of AI and scientific computing.

Di Guglielmo, Giuseppe [Fermilab]↗
DOE OSTI2025

On the effectiveness of neural operators at zero-shot weather downscaling

Machine-learning (ML) methods have shown great potential for weather downscaling. These data-driven approaches provide a more efficient alternative for producing high-resolution weather datasets and forecasts compared to physics-based numerical simulations. Neural operators, which learn solution operators for a family of partial differential equations, have shown great success in scientific ML applications involving physics-driven datasets. Neural operators are grid-resolution-invariant and are often evaluated on higher grid resolutions than they are trained on, i.e., zero-shot super-resolution. Given their promising zero-shot super-resolution performance on dynamical systems emulation, we present a critical investigation of their zero-shot weather downscaling capabilities, which is when models are tasked with producing high-resolution outputs using higher upsampling factors than are seen during training. To this end, we create two realistic downscaling experiments with challenging upsampling factors (e.g., 8x and 15x) across data from different simulations: the European Centre for Medium-Range Weather Forecasts Reanalysis version 5 (ERA5) and the Wind Integration National Dataset Toolkit. While neural operator-based downscaling models perform better than interpolation and a simple convolutional baseline, we show the surprising performance of an approach that combines a powerful transformer-based model with parameter-free interpolation at zero-shot weather downscaling. We find that this Swin-Transformer-based approach mostly outperforms models with neural operator layers in terms of average error metrics, whereas an Enhanced Super-Resolution Generative Adversarial Network-based approach is better than most models in terms of capturing the physics of the ground truth data. We suggest their use in future work as strong baselines.

17 WIND ENERGY↗