Engineering Papers⌕ Search

SEARCH · Engineering Papers

Results for “data similarity”

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 415 records · Page 23
All sourcesNASA NTRSDOE OSTI
DOE OSTI2025

SAXS Assistant: Automated SAXS analysis for structural discovery in biologics and polymeric nanoparticles

Small-angle x-ray scattering (SAXS) is a powerful technique for assessing macromolecular structure. High-throughput SAXS is limited by the time-consuming and, at times, subjective nature of SAXS data interpretation. Here, we present SAXS Assistant, a Python-based script that streamlines SAXS data analysis to extract features for machine learning (ML) and key structural parameters, including the Guinier radius of gyration (R g ), pair distance distribution function (PDDF)-derived R g , maximum particle dimension (D max ), and Kratky plots. The script builds upon BioXTAS RAW and validates reliability via Guinier/PDDF R g agreement, an important indicator of well-measured data sets. For assistance in D max estimation, a multilayer perceptron regressor was trained with 1940 data files from the Small Angle Scattering Biological Data Bank. The model achieved a test set performance R 2 = 0.90 and mean absolute error = 11.7 Å. Training exclusively with experimental data translates analyses from researchers, including experts in the field, to the ML model, which helps assess D max estimations from PDDF. Gaussian mixture model clustering was implemented to classify profiles into structural classes based on entries in the Small Angle Scattering Biological Data Bank. Users may therefore assess the similarity between experimental samples and known biomolecular shapes within the mapped repository entries. This probabilistic clustering aids in quantifying information from Kratky and generating shape-descriptive features. SAXS Assistant accelerates SAXS data analysis through enforced quality control, ML-ready outputs, and flags for low-confidence results. In addition to providing the ability to analyze large data sets at high throughput, this tool is versatile and may serve researchers in both biological and synthetic polymer research fields.

36 MATERIALS SCIENCE↗
DOE OSTI2023

Comparing storm resolving models and climates via unsupervised machine learning

Global storm-resolving models (GSRMs) have gained widespread interest because of the unprecedented detail with which they resolve the global climate. However, it remains difficult to quantify objective differences in how GSRMs resolve complex atmospheric formations. This lack of comprehensive tools for comparing model similarities is a problem in many disparate fields that involve simulation tools for complex data. To address this challenge we develop methods to estimate distributional distances based on both nonlinear dimensionality reduction and vector quantization. Our approach automatically learns physically meaningful notions of similarity from low-dimensional latent data representations that the different models produce. This enables an intercomparison of nine GSRMs based on their high-dimensional simulation data (2D vertical velocity snapshots) and reveals that only six are similar in their representation of atmospheric dynamics. Furthermore, we uncover signatures of the convective response to global warming in a fully unsupervised way. Our study provides a path toward evaluating future high-resolution simulation data more objectively.

54 ENVIRONMENTAL SCIENCES↗
DOE OSTI2025

Enhancing Unknown Waveform Detection by Learning Intra and Inter-domain Dependencies with Advanced Attention Fusion Mechanisms

Detection of unknown waveforms in mission-critical communications is a crucial area of interest for the Department of Energy (DoE). Traditional methods and recent deep learning-based approaches often assume that the training set includes all possible classes, which is impractical for detecting new waveforms. This limitation gives rise to the problem of open-set recognition (OSR), which involves correctly identifying known classes while detecting and rejecting unknown or unseen classes. To address this limitation, we propose a novel dual-domain complex-valued neural architecture that jointly processes time-domain and frequency-domain signal representations using transformer mechanisms. A transformer model is a deep learning architecture that uses self-attention mechanisms to process and learn relationships in sequential data. Our model employs a cosine similarity loss to extract domain-specific features and incorporates a transformer architecture in the latent space to weigh the importance of different features from the time and frequency domains. The transformer layer includes stacked self-attention and cross-attention modules to learn intra-domain and inter-domain dependencies, creating a more holistic signal representation. An attention-based fusion module intelligently combines the time and frequency-domain features using multi-head attention, enabling the network to learn the optimal feature for each domain in each input signal. Quantitative results demonstrate the impact of these architectural choices on overall performance, showing significant improvement after incorporating self and cross-attention modules and using complex attention fusion over simple weighted fusion. Our ongoing work will focus on addressing the limitations of threshold-based OSR methods by developing a novel generative framework that integrates a conditional diffusion probabilistic model (DPM). DPM is a generative framework that learns to synthesize complex data by reversing a gradual noising process using a neural network trained to denoise step-by-step. Our goal is to leverage the inherent strengths of DPMs for identifying unknown signals more robustly. One primary advantage of using a DPM is its ability to provide a more reliable anomaly score based on the model's reconstruction error, rather than relying solely on classifier confidence. Additionally, the iterative denoising process of DPMs makes this approach naturally resilient to low Signal-to-Noise Ratio (SNR) conditions, where traditional methods often fail. By implementing this generative framework, we aim to enhance the model's capability to accurately detect unknown waveforms and maintain performance in challenging environments.

99 - GENERAL AND MISCELLANEOUS↗
DOE OSTI2022

EUCLID: Experiments Underpinned by Computational Learning for Improvements in Nuclear Data [Slides]

EUCLID will design validation experiments optimized to resolve compensating errors and adjust nuclear data to experiments. A big part of the success of the EUCLID proposal was due to previous work supported by NCSP (MCNP, nuclear data, and NCERC capabilities). The work performed under EUCLID will similarly benefit the NCSP mission. It will lead to new MCNP capabilities, improved nuclear data and nuclear data capabilities, and new methodology and tools that will have large impact on future NCERC experiments.

73 NUCLEAR PHYSICS AND RADIATION PHYSICS↗
DOE OSTI2020

Methane Emissions from Gathering Compressor Stations in the U.S.

Using results from a nationally representative measurement campaign at 180 gathering compressor stations conducted with nine industry partners, this study estimated emissions for the U.S. gathering sector, where sector-specific emission factors have not been previously available. The study drew from a partner station population of 1705 stations—a significantly larger pool than was available for prior studies. Data indicated that whole gas emission rates from components on gathering stations were comparable to or higher than emission factors utilized by the EPA’s greenhouse gas reporting program (GHGRP) but less than emission factors used for similar components on transmission compressor stations. Field data also indicated that the national population of stations likely has a higher fraction of smaller stations, operating at lower throughput per station, than the data used to develop the per-station emission factor used in EPA’s greenhouse gas inventory (GHGI). This was the first national study to incorporate extensive activity data reported to the GHGRP, including 319 basin-level reports, covering 15,895 reported compressors. Further, combining study emission data with 2017 GHGRP activity data, the study indicated statistically lower national emissions of 1290 [1246–1342] Gg methane per year or 66% [64–69%] of current GHGI estimates, despite estimating 17% [12–22%] more stations than the 2017 GHGI (95% confidence interval). Finally, we propose a replicable method that uses GHGRP activity data to annually update GHGI gathering and boost sector emissions.

03 NATURAL GAS↗
DOE OSTI2026

Expanded Intercomparison of Nuclear Data Libraries Using Jupiter and Jupiter High-240 Experiments

There is a limited availability of plutonium experiments with sensitivity to lead in the ICSBEP (International Handbook of Evaluated Criticality Safety Benchmark Experiments) Handbook. The Jupiter and Jupiter High-240 experiments were performed at the National Criticality Experiments Research Center as a collaborative effort between Los Alamos National Laboratory and the Japan Atomic Energy Agency to assess lead void coefficients in a plutonium-lead system containing weapons- and reactor-grade plutonium, respectively. Concurrent with benchmark development, an intercomparison of calculations using different nuclear data libraries has been performed to assess the usability of the experimental data for nuclear data adjustment in a “softer-that-fast” neutron energy spectrum. Eigenvalue calculations using MCNP with the ENDF/B-VIII.0 and TENDL-2021 nuclear data libraries calculate closest to the benchmark values for Jupiter. Calculations using JENDL-5 and ENDF/B-VIII.1 match best with the Jupiter High-240 values. Lead void worth calculations using the various nuclear data libraries are all within 3σ of their respective measured values. Perturbation studies between ENDF/B-VIII.0 and ENDF/B-VIII.1 demonstrate an approximate increase in calculated eigenvalues for the Jupiter series experiments by ~240 pcm for plutonium (mostly 239 Pu) and ~120 pcm for lead accompanied by a decrease contributed by ~113 pcm for copper and ~13 pcm for stainless steel. Nuclear data sensitivities and uncertainties investigated using Whisper show slightly lower sensitivity to scatter than a lead-reflected plutonium sphere but greater sensitivity to neutron capture. The sensitivities between Jupiter and Jupiter High-240 for lead are very similar for both ENDF/B-VIII.0 and ENDF/B-VIII.1 nuclear data. These benchmarks are more sensitive to neutron capture in lead than other plutonium benchmark experiments and would be useful for both lead and 240 Pu validation. In conclusion, with the high degree of compensating effects between copper, lead, and plutonium cross sections, additional isolated Pb-Pu and Cu-Pu benchmarks would be beneficial in improving these nuclear data.

11 NUCLEAR FUEL CYCLE AND FUEL MATERIALS↗
DOE OSTI2020

An Accelerated Creep Testing Program For Nickel Based Superalloys

In order to improve the energy production within the United States, new materials are developed with the goal of increased functionality. These materials are required to perform at extreme pressures and temperatures (up to and beyond 200MPa and 650â??) as well as be sustainable beyond 105 hours. Before developing these materials into components for power generation, mechanical testing is required to better understand the behavior of these materials at various loads and temperatures. The longest of these tests is the conventional creep test which is a real time test that can beyond 105 hours depending on service conditions. Hundreds of specimens must be tested before being approved for use in power generation, but it is extremely difficult to rapidly qualify when each material is required to exceed 105 hours making testing through conventional means unfeasible. While others have designed accelerated testing methods to help qualify materials, many of these experiments use an empirical approach that requires test data to fit already existing creep data. In this study, an accelerated creep test called the stepped isostress method (SSM) is then applied to wrought Inconel 718 (IN 718) specimens. A test matrix is designed using deformation mechanism maps, time-temperature-transformation diagrams, and time-temperature-precipitation diagrams, in order to ensure that microstructurally, the accelerated creep tests fail similarly to the conventional creep tests. The data from the stepped isostress method is then applied to the McVetty and Sin-hyperbolic (Sinh) constitutive models in order to generate a creep master curve. The results from the model fits are then compared to the conventional creep data done from experiments as well as from the NIMS database for Inconel 718. Optical microscopy and scanning electron microscopy are used to compare the evolution of the microstructure from the conventional creep and accelerated creep tests. It is found that using the Sinh model on the accelerated creep data provides reasonable creep rupture predictions when compared to the scatter band of creep tests. It was also observed that the microstructural evolution between the SSM experiments and creep tests were very similar in terms of the volume fraction of precipitates and phases.

Mach, Robert↗
DOE OSTI2021

Topological network analysis of patient similarity for precision management of acute blood pressure in spinal cord injury

Background: Predicting neurological recovery after spinal cord injury (SCI) is challenging. Using topological data analysis, we have previously shown that mean arterial pressure (MAP) during SCI surgery predicts long-term functional recovery in rodent models, motivating the present multicenter study in patients. Methods: Intra-operative monitoring records and neurological outcome data were extracted (n = 118 patients). We built a similarity network of patients from a low-dimensional space embedded using a non-linear algorithm, Isomap, and ensured topological extraction using persistent homology metrics. Confirmatory analysis was conducted through regression methods. Results: Network analysis suggested that time outside of an optimum MAP range (hypotension or hypertension) during surgery was associated with lower likelihood of neurological recovery at hospital discharge. Logistic and LASSO (least absolute shrinkage and selection operator) regression confirmed these findings, revealing an optimal MAP range of 76–[104-117] mmHg associated with neurological recovery. Conclusions: We show that deviation from this optimal MAP range during SCI surgery predicts lower probability of neurological recovery and suggest new targets for therapeutic intervention. Funding: NIH/NINDS: R01NS088475 (ARF); R01NS122888 (ARF); UH3NS106899 (ARF); Department of Veterans Affairs: 1I01RX002245 (ARF), I01RX002787 (ARF); Wings for Life Foundation (ATE, ARF); Craig H. Neilsen Foundation (ARF); and DOD: SC150198 (MSB); SC190233 (MSB); DOE: DE-AC02-05CH11231 (DM).

59 BASIC BIOLOGICAL SCIENCES↗
DOE OSTI2022

A Package of Momentum and Heat Transfer Coefficients for the Stable Surface Layer Extended by New Coefficients over Sea Ice

The bulk transfer coefficients of momentum, heat, and humidity belong to the main ingredients of numerical weather prediction and climate models. They are needed for the calculation of turbulent fluxes in the surface layer and often rely on the Monin–Obukhov similarity theory requiring universal stability functions. The problem of a derivation of transfer coefficients based on different stability functions has been considered by many researchers over the years but it remains to this day. In this work, dedicated to the memory of S.S. Zilitinkevich, we also address this task, and obtain transfer coefficients from three pairs of theoretically derived stability functions suggested by Zilitinkevich and co-authors for stable conditions. Additionally, we construct non-iterative parametrizations of these transfer coefficients based on earlier work. Results are compared with state-of-the-art coefficients for land, ocean, and sea ice. The combined parametrizations form a package in a universal framework relying on a semi-analytical solution of the Monin-Obukhov similarity theory equations. A comparison with data of the Surface Heat Budget of the Arctic Ocean campaign (SHEBA) over sea ice reveals large differences between the coefficients for land conditions and the measurements over sea ice. However, two schemes of Zilitinkevich and co-authors show, after slight modification, good agreement with SHEBA although they had not been especially developed for sea ice. One pair of the modified transfer coefficients is superior and is compatible to earlier SHEBA-based parametrizations. Finally, an algorithm for practical use of all transfer coefficients in climate models is given.

54 ENVIRONMENTAL SCIENCES↗
DOE OSTI2021

High spatial resolution temperature profile measurements of solid-oxide fuel cells

Temperature gradients resulting from local electrochemical reactions, current distribution and geometry of gas flow channels in solid oxide fuel cells (SOFCs) create thermal stresses, localized thermophysical property gradients and uneven property evolution, contributing to SOFC degradation. This paper presents a new method to perform temperature measurements (up to 800°C) at high spatial resolutions to monitor the operation of SOFCs. Using femtosecond laser irradiation, distributed fiber sensors were hardened for high temperature environment applications. Distributed fiber sensors were embedded in interconnected plates using an additive manufacturing method to perform temperature measurements with 4-mm spatial resolution during the operation of a planar fuel cell. The measurement revealed the impact of various H 2 fuel concentrations and current loads have on temperature profiles of the SOFC tested. Temperature variation on the anode side was found to be less than 5°C, and 3°C on the cathode side. The measurements were compared to results from a multiphysics fuel cell performance model simulating similar conditions. These simulations predicted similar temperature gradients, indicating the experimental data obtained is reasonable. The model also predicts that the effect of the embedded sensor has on the local temperature will be minimal and that the gradient of temperature in the gas channels will be captured despite the separation between the sensor and the gas flow. Finally, the high spatial resolution data harnessed by these distributed fiber sensors provides experimental support for model-based design and optimization to improve the operational efficiency and longevity of solid oxide fuel cells and fuel cell assemblies.

25 ENERGY STORAGE↗
DOE OSTI2025

Permeate fluxes from desalination of brines and produced waters: A reactive transport modeling study

The increasing interest in the use of membrane systems to desalinate inland brackish water, agricultural drainage, and industrially produced wastewater demands improved means of predicting desalination system performance under variable feedwater compositions. The interaction among water flow, solute transport, and chemical composition in these systems impacts permeate flux evolution. Here, an established multicomponent reactive transport simulator that accounts for these coupled processes is applied to compute osmotic pressure and permeate fluxes in reverse osmosis (RO) systems. The model is first validated by predicting permeate fluxes for a set of benchtop crossflow experiments subject to a range of feed flow rates and compositions, under fouling and non-fouling conditions. Results compare favorably with measured data that show that solutions with similar total dissolved solids concentrations but different compositions result in different permeate fluxes. The model is then applied to predict permeate fluxes from the desalination of produced waters using a commercial spiral wound RO module. For NaCl-dominant brines, at total dissolved salt concentrations (TDS) below about 70 g/L, permeate fluxes are inversely proportional to water mole fraction as the latter is a reasonable approximation of water activity (i.e. ideal mixing). In the case of Ca–Cl-, Na–CO3- and Na–SO4-dominant brines below about 70 g/L TDS, this relationship does not hold as well and tends to overpredict osmotic pressure and thus underpredict permeate fluxes. However, the opposite becomes true at higher TDS values for typical produced waters. The scaling potential of these waters is also computed by allowing the precipitation of minerals above their saturation limit on the RO membrane. This work demonstrates how reactive transport models developed for the analysis of waters from geological systems can be extended to improve process design, optimization, and control in desalination systems from produced waters and beyond.

Molins, Sergi↗
DOE OSTI2023

Validation of Octane Hyperboosting Phenomenon in Prenol and Structurally Related Olefinic Alcohols

Hyperboosting is a recently discovered phenomenon in which the research octane number (RON) of a blend is higher than both the neat blendstock and the neat fuel it was blended into. RON is a measure of a fuel's resistance to knock, and knock is a cause of engine inefficiency. Blends which exhibit hyperboosting are blends in which an overall improvement in engine efficiency may be expected. The first discovery of hyperboosting came from blending experiments in which prenol was blended into several different base fuels. Here, ignition delay time (IDT) measurements taken using a commercially available constant volume combustion chamber called the Advanced Fuel Ignition Delay Analyzer (AFIDA) are presented. The data show that some prenol blends have longer IDTs (lower reactivity) than either neat prenol or the base fuel, providing further evidence of hyperboosting. Additionally, more blending data is presented in which the base fuel is varied, which allows for a better understanding of hyperboosting sensitivity to chemical classes. The data indicate that aromatics may inhibit, and branched alkanes may enhance the magnitude of hyperboosting observed. Enthalpy of vaporization estimates are also given for several molecules which are blended into a 4-component surrogate. These estimates are derived from Equation of State simulations and reveal that there is no observable correlation between hyperboosting and enthalpy of vaporization. Blending data for molecules which share structural similarities with prenol are also presented. Structure property relationships are suggested, in which the double bond motif of prenol seems to play an important role in hyperboosting. This may help to understand hyperboosting and its underlying mechanism. Lastly, blending curves of surrogate blends with prenol experienced hyperboosting under lean (Homogeneous Charged Compression Ignition-HCCI) operating conditions, which validates that hyperboosting is not an artifact of the octane test methods, but inherent to the properties of prenol.

ADVANCED PROPULSION SYSTEMS↗
DOE OSTI2022

Development and application of a uranium mononitride (UN) potential: Thermomechanical properties and Xe diffusion

We report that atomic-scale modeling of thermophysical and defect properties of uranium mononitride (UN) plays an important role in establishing a better understanding and improved models of UN fuel performance. Having an accurate interatomic potential is crucial for generating reliable data at finite temperatures using molecular dynamic simulations. We report a new interatomic potential for UN, based on a combination of many-body and pairwise interactions, a simple form that we later show could be easily adapted to include Xe-U and Xe-N interactions, i.e., generating a U-N-Xe interatomic potential. The potential was fitted to experimental thermal expansion and single crystal elastic constants, as well as Frenkel, Schottky, anti-Schottky, and antisite pair reaction energies from density functional theory (DFT) calculations. Using the potential, we successfully reproduced experimental lattice parameters, thermal expansion, single crystal elastic constants, and temperature dependent heat capacity. The potential also performs reasonably well in reproducing the energy of the aforementioned stoichiometric defect reactions and defect migration barriers calculated using DFT. However, the potential underestimates the energy difference between the tetrahedral and dumbbell uranium interstitials, and a more complex potential form might be needed to overcome this issue. The potential was also used to predict UN single crystal elastic constants and elastic properties at different temperatures, showing that UN becomes softer and more compressible with increasing temperature. We also compare our potential against literature data from previous empirical potentials, demonstrating similar or better behavior depending on the property of interest. To enable the simulation of Xe in UN a Buckingham potential has been fitted to DFT-derived Xe incorporation energies. The potential was then used to determine the activation energy for Xe diffusion due to various Xe-containing defects, with {XeU : VU} exhibiting the lowest activation energy.

37 INORGANIC, ORGANIC, PHYSICAL, AND ANALYTICAL CH↗
DOE OSTI2023

Comparing Adjoint Waveform Tomography Models of California Using Different Starting Models

Abstract Adjoint waveform tomography (AWT) sits at the cutting edge of seismic tomography on local, regional, and global scales. However, the choice in starting model may have a significant impact on the final inversion results. In this paper, we present 3 AWT models of California that are based on different starting models. We chose three models that were inverted at different scales: SPiRaL, a global travel‐time tomography model (Simmons et al., 2021, 10.1093/gji/ggab277 ), CSEM_NA, a regional adjoint tomography model of North America and the North Atlantic (Krischer et al., 2018, 10.1029/2017JB015289 ), and WUS256, a regional adjoint tomography model of the western US (Rodgers et al., 2022, https://doi.org/10.1029/2022JB024549 ). We then inverted three AWT models using the same source and receiver set. We ran each model over three period bands: 30–100 s, 25–100 s, and 20–80 s. Once the iterations were finalized, we used five methods of testing model similarity in both the model and data space. We conclude that the choice of starting model has a minimal impact on long wavelength models if an appropriate multi‐scale inversion approach is used.

58 GEOSCIENCES↗
DOE OSTI2022

A substrate-induced gating mechanism is conserved among Gram-positive IgA1 metalloproteases

The mucosal adaptive immune response is dependent on the production of IgA antibodies and particularly IgA1, yet opportunistic bacteria have evolved mechanisms to specifically block this response by producing IgA1 proteases (IgA1Ps). Our lab was the first to describe the structures of a metal-dependent IgA1P (metallo-IgA1P) produced from Gram-positive Streptococcus pneumoniae both in the absence and presence of its IgA1 substrate through cryo-EM single particle reconstructions. This prior study revealed an active-site gating mechanism reliant on substrate-induced conformational changes to the enzyme that begged the question of whether such a mechanism is conserved among the wider Gram-positive metallo-IgA1P subfamily of virulence factors. Here, we used cryo-EM to characterize the metallo-IgA1P of a more distantly related family member from Gemella haemolysans, an emerging opportunistic pathogen implicated in meningitis, endocarditis, and more recently bacteremia in the elderly. While the substrate-free structures of these two metallo-IgA1Ps exhibit differences in the relative starting positions of the domain responsible for gating substrate, the enzymes have similar domain orientations when bound to IgA1. Together with biochemical studies that indicate these metallo-IgA1Ps have similar binding affinities and activities, these data indicate that metallo-IgA1P binding requires the specific IgA1 substrate to open the enzymes for access to their active site and thus, largely conform to an “induced fit” model.

59 BASIC BIOLOGICAL SCIENCES↗
DOE OSTI2023

Star formation rate and stellar mass calibrations based on infrared photometry and their dependence on stellar population age and extinction

The stellar mass (M $\star$ ) and the star formation rate (SFR) are among the most important features that characterize galaxies. Measuring these fundamental properties accurately is critical for understanding the present state of galaxies, their history, and future evolution. Infrared (IR) photometry is widely used to measure the M $\star$ and SFR of galaxies because the near-IR traces the continuum emission of the majority of their stellar populations (SPs), and the mid/far-IR traces the dust emission powered by star-forming activity. This work explores the dependence of the IR emission of galaxies on their extinction, and the age of their SPs. It aims to provide accurate and precise IR-photometry SFR and M $\star$ calibrations that account for SP age and extinction while providing quantification of their scatter. We used the CIGALE spectral energy distribution (SED) fitting code to create model SEDs of galaxies with a wide range of star formation histories, dust content, and interstellar medium properties. We fit the relations between M $\star$ and SFR with IR and optical photometry of the model-galaxy SEDs with the Markov chain Monte Carlo (MCMC) method. As an independent confirmation of the MCMC fitting method, we performed a machine-learning random forest (RF) analysis on the same data set. The RF model yields similar results to the MCMC fits, thus validating the latter. This work provides calibrations for the SFR using a combination of the WISE bands 1 and 3, or the JWST NIR-F200W and MIRI-F2100W. It also provides mass-to-light ratio calibrations based on the WISE band-1, the JWST NIR-F200W, and the optical u - r or g - r colors. These calibrations account for the biases attributed to the SP age, while they are given in the form of extinction-dependent and extinction-independent relations. The proposed calibrations show robust estimations while minimizing the scatter and biases throughout a wide range of SFRs and stellar masses. The SFR calibration offers better results, especially in dust-free or passive galaxies where the contributions of old SPs or biases from the lack of dust are significant. Similarly, the M $\star$ calibration yields significantly better results for dusty and high-SFR galaxies where dust emission can otherwise bias the estimations.

79 ASTRONOMY AND ASTROPHYSICS↗
DOE OSTI2025

Combining physics-based and data-driven models for quantitatively accurate plasma profile prediction that extrapolates well; with application to DIII-D, AUG, and ITER tokamaks

For design, scenario planning, and control, ITER and all other envisioned tokamaks rely on a variety of statistical and physics-based models to extrapolate to unseen regimes; most notably from low plasma current to high. A 'meta-learning' methodology for combining the accuracy of data-driven models with the generalizability of physics-based models is described and tested, yielding a 5–10 percent improvement in performance beyond either alone for the task of extrapolating time-dependent plasma profile prediction from low- to high- plasma current DIII-D tokamak discharges. Meanwhile, it is shown that both machine learning models extrapolated far-distribution and state-of-the-art 'physics-based' profile predictors fare worse than merely assuming plasma profiles do not change from their initial values. Finally, a variety of other mechanisms for helping data-driven models generalize—transfer learning, adding contextual information from physics simulators, and adding data from the ASDEX Upgrade tokamak—are attempted for similar extrapolation tasks but, in the methodology used in this paper, yield no significant improvement beyond simple data-driven models. Results are summarized in figures 15 and 16.

70 PLASMA PHYSICS AND FUSION TECHNOLOGY↗
DOE OSTI2024

Harnessing large language models’ zero-shot and few-shot learning capabilities for regulatory research

Abstract Large language models (LLMs) are sophisticated AI-driven models trained on vast sources of natural language data. They are adept at generating responses that closely mimic human conversational patterns. One of the most notable examples is OpenAI's ChatGPT, which has been extensively used across diverse sectors. Despite their flexibility, a significant challenge arises as most users must transmit their data to the servers of companies operating these models. Utilizing ChatGPT or similar models online may inadvertently expose sensitive information to the risk of data breaches. Therefore, implementing LLMs that are open source and smaller in scale within a secure local network becomes a crucial step for organizations where ensuring data privacy and protection has the highest priority, such as regulatory agencies. As a feasibility evaluation, we implemented a series of open-source LLMs within a regulatory agency’s local network and assessed their performance on specific tasks involving extracting relevant clinical pharmacology information from regulatory drug labels. Our research shows that some models work well in the context of few- or zero-shot learning, achieving performance comparable, or even better than, neural network models that needed thousands of training samples. One of the models was selected to address a real-world issue of finding intrinsic factors that affect drugs' clinical exposure without any training or fine-tuning. In a dataset of over 700 000 sentences, the model showed a 78.5% accuracy rate. Our work pointed to the possibility of implementing open-source LLMs within a secure local network and using these models to perform various natural language processing tasks when large numbers of training examples are unavailable.

Biochemistry & Molecular Biology↗