Search NASA⌕ Search

SEARCH · Search NASA

Results for “Linear Regression”

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 109 records · Page 6

Regression Bias in using Solar Wind Measurements

Simultaneous solar wind measurements from the solar wind monitors, WIND and ACE, differ due to the spatial and temporal structure of the solar wind. Correlation studies that use these measurements as input may infer an incorrect correlation due to uncertainties arising from this spatial and temporal structure, especially at extreme and rare solar wind values. In particular, regression analysis will lead to a regression function whose slope is biased towards the mean value of the measurement parameter. This article demonstrates this regression bias by comparing simultaneous ACE and WIND solar wind measurements. A non-linear regression analysis between them leads to a perception of underestimation of extreme values of one measurement on average over the other. Using numerical experiments, we show that popular regression analysis techniques such as linear least-squares, orthogonal least-squares, and non-linear regression are not immune to this bias. Hence while using solar wind parameters as an independent variable in a correlation or regression analysis, random uncertainty in the independent variable can create unintended biases in the response of the dependent variable. More generally, the regression to the mean effect can impact both event-based, statistical studies of magnetospheric response to solar wind forcing.

magnetosphere↗

Global Variability in Sonic Boom Exposure due to Macroscopic Effects

Supersonic flight over land has been prohibited since 1973 due to the loudness of sonic booms. NASA is building the X-59 aircraft as part of its Quesst mission to demonstrate low-loudness shaped sonic booms, or “sonic thumps.” The Quesst mission will gather human perception data via a series of community noise surveys across the USA. The noise dose and perceptual response data will be provided to the International Civil Aviation Organization (ICAO) and the Federal Aviation Administration for use in determining potential future supersonic aircraft noise certification standards, effectively changing the prohibition from a speed limit to a noise limit. These noise regulations must be globally effective, as long travel distances see the largest benefit to supersonic flight. The state of the atmosphere through which a sonic boom travels affects the size of the region exposed to sound, the “carpet width” (CW), as well as the loudness. The focus of this dissertation is to understand and quantify the expected loudness and CW of sonic booms due to the macroscopic atmospheric effects around the world. A pair of large-scale propagation simulation studies were conducted using the NASA PCBoom code to compare predicted sonic boom loudness and CW statistics first across the USA and then across the world. For the USA study, near-field data of the X-59 in steady cruise was propagated at 4 cardinal headings at 138 locations through 5 years of Climate Forecast System Version 2 (CFSv2) atmospheric profiles. Results of a bootstrap forest predictor screening model indicated the importance of climate zone, latitude, ground elevation, season, and heading. It also noted the unimportance of time of day for predicting loudness and CW. The data is visualized in aggregate, and then broken out geographically, by season and heading, and by climate zone. Multiple linear regression models were fit to the data from the 138 locations so that estimates of the loudness and CW can be produced anywhere in the US. The results can aid in planning when and where to fly the X-59. For the global study, near-field data from three aircraft, the X-59 in a quiet and loud configuration, B-58, and Concorde, were propagated at four cardinal headings through data from three atmospheric models, the CFSv2, the Global Forecast System (GFS), and the ECMWF Reanalysis Version 5 (ERA5), at 100 global locations over 1 year. Results of a bootstrap forest predictor screening model indicated the importance of climate zone, ground elevation, season, and heading. Similar to the US study, the model indicated time of day was not an important predictor. The model also indicated that choice of weather model was not important, so the atmospheric model data are effectively interchangeable. The ERA5 model was chosen for use in an extension of the study to include 18 additional locations to ensure sampling of every climate zone. Loudness and CW results are shown in aggregate, and split geographically and by heading, season, and climate. Multiple linear regression models were fit to the data from the 118 locations so that estimates of loudness and CW can be produced around the world. N-waves and shaped booms did not have the same global variability. Koppen-Geiger climate zones were used as the climate zone definition for the global study. These are available as present-day and future climate projections. Making use of the multiple linear regression models, the future climate zones were input to estimate the effect of the changing climate on sonic boom loudness and CW. Results indicate that a changing climate would have little impact on the effectiveness of noise regulations.

X-59↗

Fundamental Analysis of the Linear Multiple Regression Technique for Quantification of Water Quality Parameters from Remote Sensing Data

Constituents with linear radiance gradients with concentration may be quantified from signals which contain nonlinear atmospheric and surface reflection effects for both homogeneous and non-homogeneous water bodies provided accurate data can be obtained and nonlinearities are constant with wavelength. Statistical parameters must be used which give an indication of bias as well as total squared error to insure that an equation with an optimum combination of bands is selected. It is concluded that the effect of error in upwelled radiance measurements is to reduce the accuracy of the least square fitting process and to increase the number of points required to obtain a satisfactory fit. The problem of obtaining a multiple regression equation that is extremely sensitive to error is discussed.

Whitlock, C. H., III↗

Automated and efficient local adaptive regression for principal component-based reduced-order modeling of turbulent reacting flows

Principal Component Analysis can be used to reduce the cost of Computational Fluid Dynamics simulations of turbulent reacting flows by reducing the dimensionality of the transported variables through projection of the thermochemical state onto a lower-dimensional manifold. However, because of the nonlinearity of the principal component source terms, nonlinear regression techniques must be utilized for the source terms in terms of the principal components. Unfortunately, widely available and utilized nonlinear regression techniques can have prohibitive computational requirements and/or accuracy that is highly dependent on user experience in ad hoc tuning of model architecture and hyperparameters. Here, in this work, a new nonlinear regression approach is proposed that is both computationally efficient and automated so does not require any user input. The approach is evaluated through a priori prediction of principal component source terms using data from a Direct Numerical Simulation of a turbulent nonpremixed n-heptane/air jet flame. In particular, the proposed framework consists of local regressions whose complexity is adapted according to the local nonlinearity of the data: local linear regression when accurate enough and local Artificial Neural Networks when nonlinear regression is required. The number of local clusters for local regression is determined automatically using the Davies-Bouldin index. In addition, Bayesian optimization is utilized for model training (i.e., to select the best architectures and hyperparameters of the nonlinear regressions in an unsupervised fashion), eliminating ad hoc hand-tuning and/or expensive grid searches. Overall, compared to a single, global neural network, the new local adaptive regression approach is shown to have comparable accuracy but 69% less training time due to the utilization of local linear regression and faster training of local neural networks.

42 ENGINEERING↗

Sensor selection and tool wear prediction with data‐driven models for precision machining

Abstract Estimation of tool wear in precision machining is vital in the traditional subtractive machining industry to reduce processing cost, improve manufacturing efficiency and product quality. In this vein, fusion of time and frequency‐domain features of commonly sensed signals can provide an early indication of tool wear and improve its prediction accuracy for prognostics and health management. This paper presents a data‐driven methodology and a complete tool chain for the inference of precision machining tool wear from fused machine measurements, such as cutting force, power, audio and vibration signals, and quantify the usefulness of each measurement. Indicators of tool wear are extracted from time‐domain signal statistics, frequency‐domain analysis, and time‐frequency domain analysis. Correlation coefficients between the extracted features (indicators) and the tool wear are used to select the most informative features. Principal Component Analysis and Partial Least‐Squares are used to reduce the dimensionality of the feature space. Regression models, including linear regression, support vector regression, Decision tree regression, neural network regression and Gaussian process regression, are used to predict the tool wear using data from a Haas milling machine performing spiral boss face milling. The performance of the regression models based on subsets of sensors validates the preliminary estimates about the saliency of the sensors. The experimental results show that the proposed methods can predict the machine tool wear precisely, with readily available sensor measurements. Neural network and Gaussian process regression were able to achieve good estimates of tool wear at different machine operating conditions. The most informative signal in predicting tool wear was shown to be the vibration signal. Time‐frequency domain features were the most informative features among the combination of features of three domains. In addition, using partial least squares components extracted from the original features of signals led to higher prediction accuracy.

Han, Seulki↗

Assessing the large-scale drivers of precipitation in the northeastern United States via linear orthogonal decomposition

This study examines the linear orthogonal modes associated with monthly precipitation in the northeastern United States, from CESM1 LENS (35 ensemble members, 1979–2005) and two reanalysis datasets (ERA5, 1950–2018 and NOAA-CIRES-DOE 20CRv3, 1950–2015). Calendar months are aggregated together, and any linear trends in data are removed. Using region-averaged precipitation anomaly time series and monthly anomalies for several global 2D atmospheric fields, a linear orthogonal decomposition method is implemented to iteratively extract time series (based on field and geographic location) of absolute maximum correlation. Linear modes associated with this method are then projected onto the full set of 2D fields to provide physical insight into the mechanisms involved in generating precipitation. In this region, the first mode is associated with vapor transport from the Atlantic seaboard, the second mode is characterized by westward vapor transport associated with extratropical cyclones, and the third mode captures vapor transport from the Gulf of Mexico during the fall and winter. However, the third mode is less robust in the spring and summer. Results are generally consistent across the datasets, and applying multiple linear regression with the linear modes to predict the precipitation anomalies produces R-squared values of around 0.54–0.65 for CESM1 LENS, and around 0.58–0.88 for reanalysis, with the lowest values generally in the spring and late summer. The influence of low-frequency climate variability on the modes is considered for CESM1 LENS, and the modes in late winter can be predicted with some success via a combination of several, prominent large-scale teleconnection patterns.

54 ENVIRONMENTAL SCIENCES↗

Predicting Future Hourly Residential Electrical Consumption: A Machine Learning Case Study

Whole building input models for energy simulation programs are frequently created in order to evaluate specific energy savings potentials. They are also often utilized to maximize cost-effective retrofits for existing buildings as well as to estimate the impact of policy changes toward meeting energy savings goals. Traditional energy modeling suffers from several factors, including the large number of inputs required to characterize the building, the specificity required to accurately model building materials and components, simplifying assumptions made by underlying simulation algorithms, and the gap between the as-designed and as-built building. Prior works have attempted to mitigate these concerns by using sensor-based machine learning approaches to model energy consumption. However, a majority of these prior works focus only on commercial buildings. The works that focus on modeling residential buildings primarily predict monthly electrical consumption, while commercial models predict hourly consumption. This means there is not a clear indicator of which techniques best model residential consumption, since these methods are only evaluated using low-resolution data. We address this issue by testing seven different machine learning algorithms on a unique residential data set, which contains 140 different sensors measurements, collected every 15 minutes. In addition, we validate each learner's correctness on the ASHRAE Great Energy Prediction Shootout, using the original competition metrics. Our validation results confirm existing conclusions that Neural Network-based methods perform best on commercial buildings. However, the results from testing our residential data set show that Feed Forward Neural Networks, Support Vector Regression (SVR), and Linear Regression methods perform poorly, and that Hierarchical Mixture of Experts (HME) with Least Squares Support Vector Machines (LS-SVM) performs best - a technique not previously applied to this domain.

32 ENERGY CONSERVATION, CONSUMPTION, AND UTILIZATI↗

Short-term effects of air temperature on plasma metabolite concentrations in patients undergoing cardiac catheterization

Background: Epidemiological studies have shown associations between air temperature and cardiovascular health outcomes. Metabolic dysregulation might also play a role in the development of cardiovascular disease. Objectives: To investigate short-term temperature effects on metabolites related to cardiovascular disease. Methods: Concentrations of 45 acylcarnitines, 15 amino acids, ketone bodies and total free fatty acids were available in 2869 participants from the CATHeterization GENetics cohort recruited at the Duke University Cardiac Catheterization Clinic (Durham, NC) between 2001 and 2007. Ten metabolites were selected based on quality criteria and cluster analysis. Daily averages of meteorological variables were obtained from the North American Regional Reanalysis project. Immediate, lagged, and cumulative temperature effects on metabolite concentrations were analyzed using (piecewise) linear regression models. Results: Linear temperature effects were found for glycine, C16-OH:C14:1-DC, and aspartic acid/asparagine. A 5 °C increase in temperature was associated with a 1.8% [95%-confidence interval: 0.3%; 3.3%] increase in glycine (5-day average), a 3.2% [0.1%; 6.3%] increase in C16-OH:C14:1-DC (lag of four days), and a −1.4% [−2.4%; −0.3%] decrease in aspartic acid/asparagine (lag of two days). Non-linear temperature effects were observed for alanine and total ketone bodies with breakpoint of 4 °C and 20 °C, respectively. Both a 5 °C decrease in temperature on colder days (<4 °C)and a 5 °C increase in temperature on warmer days (≥4 °C) were associated with a four day delayed increase in alanine by 6.6% [11.7; 1.8%] and 1.9% [0.3%; 3.4%], respectively. For ketone bodies we found immediate (0-day lag) increases of 4.2% [−0.5%; 9.1%] and 12.3% [0.1%; 26.0%] associated with 5 °C decreases on colder (<20 °C) days and 5 °C increases on warmer days (≥20 °C), respectively. Conclusions: We observed multiple effects of air temperature on metabolites several of which are reported to be involved in cardiovascular disease. Our findings might help to understand the link between air temperature and cardiovascular disease. - Highlights: • Certain metabolites are assumed to be novel biomarkers for cardiovascular disease. • First study investigating associations between air temperature and metabolites. • Short-term effects of temperature on amino acids, ketone bodies and acylcarnitines. • Our findings may help to understand the link between temperature and cardiovascular disease.

60 APPLIED LIFE SCIENCES↗

Attitude towards the incorporation of the selective collection of biowaste in a municipal solid waste management system. A case study

Highlights: • Attitude towards incorporating biowaste selective collection is analysed. • Willingness to participate and to pay in biowaste selective collection is obtained. • Socioeconomic aspects affecting WtParticipate and WtPay are identified. - Abstract: European waste legislation has been encouraging for years the incorporation of selective collection systems for the biowaste fraction. European countries are therefore incorporating it into their current municipal solid waste management (MSWM) systems. However, this incorporation involves changes in the current waste management habits of households. In this paper, the attitude of the public towards the incorporation of selective collection of biowaste into an existing MSWM system in a Spanish municipality is analysed. A semi-structured telephone interview was used to obtain information regarding aspects such as: level of participation in current waste collection systems, willingness to participate in selective collection of biowaste, reasons and barriers that affect participation, willingness to pay for the incorporation of the selective collection of biowaste and the socioeconomic characteristics of citizens who are willing to participate and pay for selective collection of biowaste. The results showed that approximately 81% of the respondents were willing to participate in selective collection of biowaste. This percentage would increase until 89% if the Town Council provided specific waste bins and bags, since the main barrier to participate in the new selective collection system is the need to use specific waste bin and bags for the separation of biowaste. A logit response model was applied to estimate the average willingness to pay, obtaining an estimated mean of 7.5% on top of the current waste management annual tax. The relationship of willingness to participate and willingness to pay for the implementation of this new selective collection with the socioeconomic variables (age, gender, size of the household, work, education and income) was analysed. Chi-square independence tests and binary logistic regression was used for willingness to participate, not being obtained any significant relationship. Chi-square independence tests, ordinal logistic regression and ordinary linear regression was applied for willingness to pay, obtaining statistically significant relationship for most of the socioeconomic variables.

12 MANAGEMENT OF RADIOACTIVE AND NON-RADIOACTIVE W↗

Structured chaos in a devil's staircase of the Josephson junction

The phase dynamics of Josephson junctions (JJs) under external electromagnetic radiation is studied through numerical simulations. Current-voltage characteristics, Lyapunov exponents, and Poincaré sections are analyzed in detail. It is found that the subharmonic Shapiro steps at certain parameters are separated by structured chaotic windows. By performing a linear regression on the linear part of the data, a fractal dimension of D = 0.868 is obtained, with an uncertainty of ±0.012. The chaotic regions exhibit scaling similarity, and it is shown that the devil's staircase of the system can form a backbone that unifies and explains the highly correlated and structured chaotic behavior. These features suggest a system possessing multiple complete devil's staircases. The onset of chaos for subharmonic steps occurs through the Feigenbaum period doubling scenario. Universality in the sequence of periodic windows is also demonstrated. Finally, the influence of the radiation and JJ parameters on the structured chaos is investigated, and it is concluded that the structured chaos is a stable formation over a wide range of parameter values.

71 CLASSICAL AND QUANTUM MECHANICS, GENERAL PHYSIC↗

Low thyroid function is not associated with an accelerated deterioration in renal function

Abstract Background Chronic kidney disease (CKD) is frequently accompanied by thyroid hormone dysfunction. It is currently unclear whether these alterations are the cause or consequence of CKD. This study aimed at studying the effect of thyroid hormone alterations on renal function in cross-sectional and longitudinal analyses in individuals from all adult age groups. Methods Individual participant data (IPD) from 16 independent cohorts having measured thyroid stimulating hormone, free thyroxine levels and creatinine levels were included. Thyroid hormone status was defined using clinical cut-off values. Estimated glomerular filtration rates (eGFR) were calculated by means of the four-variable Modification of Diet in Renal Disease (MDRD) formula. For this IPD meta-analysis, eGFR at baseline and eGFR change during follow-up were computed by fitting linear regression models and linear mixed models in each cohort separately. Effect estimates were pooled using random effects models. Results A total of 72 856 individuals from 16 different cohorts were included. At baseline, individuals with overt hypothyroidism (n = 704) and subclinical hypothyroidism (n = 3356) had a average (95% confidence interval) −4.07 (−6.37 to −1.78) and −2.40 (−3.78 to −1.02) mL/min/1.73 m2 lower eGFR as compared with euthyroid subjects (n = 66 542). In (subclinical) hyperthyroid subjects (n = 2254), average eGFR was 3.01 (1.50–4.52) mL/min/1.73 m2 higher. During 329 713 patient years of follow-up, eGFR did not decline more rapidly in individuals with low thyroid function compared with individuals with normal thyroid function. Conclusions Low thyroid function is not associated with a deterioration of renal function. The cross-sectional association may be explained by renal dysfunction causing thyroid hormone alterations.

Meuwese, Christiaan L.↗

Monitoring radiation use in cardiac fluoroscopy imaging procedures

Purpose: Timely identification of systematic changes in radiation delivery of an imaging system can lead to a reduction in risk for the patients involved. However, existing quality assurance programs involving the routine testing of equipment performance using phantoms are limited in their ability to effectively carry out this task. To address this issue, the authors propose the implementation of an ongoing monitoring process that utilizes procedural data to identify unexpected large or small radiation exposures for individual patients, as well as to detect persistent changes in the radiation output of imaging platforms. Methods: Data used in this study were obtained from records routinely collected during procedures performed in the cardiac catheterization imaging facility at St. Andrew's War Memorial Hospital, Brisbane, Australia, over the period January 2008-March 2010. A two stage monitoring process employing individual and exponentially weighted moving average (EWMA) control charts was developed and used to identify unexpectedly high or low radiation exposure levels for individual patients, as well as detect persistent changes in the radiation output delivered by the imaging systems. To increase sensitivity of the charts, we account for variation in dose area product (DAP) values due to other measured factors (patient weight, fluoroscopy time, and digital acquisition frame count) using multiple linear regression. Control charts are then constructed using the residual values from this linear regression. The proposed monitoring process was evaluated using simulation to model the performance of the process under known conditions. Results: Retrospective application of this technique to actual clinical data identified a number of cases in which the DAP result could be considered unexpected. Most of these, upon review, were attributed to data entry errors. The charts monitoring the overall system radiation output trends demonstrated changes in equipment performance associated with relocation of the equipment to a new department. When tested under simulated conditions, the EWMA chart was capable of detecting a sustained 15% increase in average radiation output within 60 cases (<1 month of operation), while a 33% increase would be signaled within 20 cases. Conclusions: This technique offers a valuable enhancement to existing quality assurance programs in radiology that rely upon the testing of equipment radiation output at discrete time frames to ensure performance security.

60 APPLIED LIFE SCIENCES↗

Radar estimates of soil moisture over the Konza Prairie

Radar-backscatter measurements were made to estimate soil moisture. The helicopter-mounted radar was flown along selected transects that coincided with soil-moisture measurements. The radar operated at microwave frequencies of 5.3 and 9.6 GHz and at selected incidence angles between 0 and 60 degrees. Vertical polarization was used for two days and horizontal polarization was used for three days. The scattering-coefficient data from different days were grouped by frequency and antenna angles and then related to soil moisture along the flight paths using linear regression. A measure of linearity for the regression ranged between 0.9 and 0.5. The larger coefficients were for X-band measurements made at large antenna-incidence angles, and the smaller coefficients were for C-band measurements made at incidences angles near vertical.

Gogineni, S.↗

Forecasting Lightning at Kennedy Space Center/Cape Canaveral Air Force Station, Florida

The Applied Meteorology Unit (AMU) developed a set of statistical forecast equations that provide a probability of lightning occurrence on Kennedy Space Center (KSC) I Cape Canaveral Air Force Station (CCAFS) for the day during the warm season (May September). The 45th Weather Squadron (45 WS) forecasters at CCAFS in Florida include a probability of lightning occurrence in their daily 24-hour and weekly planning forecasts, which are briefed at 1100 UTC (0700 EDT). This information is used for general scheduling of operations at CCAFS and KSC. Forecasters at the Spaceflight Meteorology Group also make thunderstorm forecasts for the KSC/CCAFS area during Shuttle flight operations. Much of the current lightning probability forecast at both groups is based on a subjective analysis of model and observational data. The objective tool currently available is the Neumann-Pfeffer Thunderstorm Index (NPTI, Neumann 1971), developed specifically for the KSCICCAFS area over 30 years ago. However, recent studies have shown that 1-day persistence provides a better forecast than the NPTI, indicating that the NPTI needed to be upgraded or replaced. Because they require a tool that provides a reliable estimate of the daily thunderstorm probability forecast, the 45 WS forecasters requested that the AMU develop a new lightning probability forecast tool using recent data and more sophisticated techniques now possible through more computing power than that available over 30 years ago. The equation development incorporated results from two research projects that investigated causes of lightning occurrence near KSCICCAFS and over the Florida peninsula. One proved that logistic regression outperformed the linear regression method used in NPTI, even when the same predictors were used. The other study found relationships between large scale flow regimes and spatial lightning distributions over Florida. Lightning, probabilities based on these flow regimes were used as candidate predictors in the equation development. Fifteen years (1 989-2003) of warm season data were used to develop the forecast equations. The data sources included a local network of cloud-to-ground lightning sensors called the Cloud-to-Ground Lightning Surveillance System (CGLSS), 1200 UTC Florida synoptic soundings, and the 1000 UTC CCAFS sounding. Data from CGLSS were used to determine lightning occurrence for each day. The 1200 UTC soundings were used to calculate the synoptic-scale flow regimes and the 1000 UTC soundings were used to calculate local stability parameters, which were used as candidate predictors of lightning occurrence. Five logistic regression forecast equations were created through careful selection and elimination of the candidate predictors. The resulting equations contain five to six predictors each. Results from four performance tests indicated that the equations showed an increase in skill over several standard forecasting methods, good reliability, an ability to distinguish between non-lightning and lightning days, and good accuracy measures and skill scores. Given the overall good performance the 45 WS requested that the equations be transitioned to operations and added to the current set of tools used to determine the daily lightning probability of occurrence.

Lambert, Winfred↗

Exploring Flooded Fraction Prediction through Machine Learning Models Focusing on Medical Infrastructure in the Southeast U.S. Coastal Areas

Rising sea levels due to climate change increasingly threaten medical infrastructure through flooding. This study develops machine learning models to predict flood exposure for 11,508 medical facilities in the southeastern coastal regions of the United States by integrating datasets including meteorological, hydrological, topographic, and geological data, the Natural Risk Index, and historical flood records from NASA, HIFLD, and FEMA. Six regression models, namely Linear Regression, Support Vector Regression, Random Forest, k-Nearest Neighbors, XGBoost, and Artificial Neural Networks, are trained using 16 explanatory variables identified through literature review and correlation analysis. Data preprocessing employs the SMOGN for class imbalance and Winsorization for outliers. Model performance is evaluated using MAE, MSE, and RMSE, with Random Forest and XGBoost models achieving the highest performance (MSE of 2.58e-5 and 3.69e-5, respectively). This multifactorial approach allows the models to capture complex flood-influencing relationships, enhancing adaptability and performance across geographic regions. Future work focuses on expanding across the U.S. and developing a near real-time flood monitoring system.

Jihoon Chung↗

Parameter inference from event ensembles and the top-quark mass

One of the key tasks of any particle collider is measurement. In practice, this is often done by fitting data to a simulation, which depends on many parameters. Sometimes, when the effects of varying different parameters are highly correlated, a large ensemble of data may be needed to resolve parameter-space degeneracies. An important example is measuring the top-quark mass, where other physical and unphysical parameters in the simulation must be profiled when fitting the top-quark mass parameter. We compare four different methodologies for top-quark mass measurement: a classical histogram fit similar to one commonly used in experiment augmented by soft-drop jet grooming; a 2D profile likelihood fit with a nuisance parameter; a machine-learning method called DCTR; and a linear regression approach, either using a least-squares fit or with a dense linearly-activated neural network. Despite the fact that individual events are totally uncorrelated, we find that the linear regression methods work most effectively when we input an ensemble of events sorted by mass, rather than training them on individual events. Although all methods provide robust extraction of the top-quark mass parameter, the linear network does marginally best and is remarkably simple. For the top study, we conclude that the Monte-Carlo-based uncertainty on current extractions of the top-quark mass from LHC data can be reduced significantly (by perhaps a factor of 2) using networks trained on sorted event ensembles. More generally, machine learning from ensembles for parameter estimation has broad potential for collider physics measurements.

72 PHYSICS OF ELEMENTARY PARTICLES AND FIELDS↗

Shock Hugoniot calculations using on-the-fly machine learned force fields with ab initio accuracy

We present a framework for computing the shock Hugoniot using on-the-fly machine learned force field (MLFF) molecular dynamics simulations. In particular, we employ an MLFF model based on the kernel method and Bayesian linear regression to compute the free energy, atomic forces, and pressure, in conjunction with a linear regression model between the internal and free energies to compute the internal energy, with all training data generated from Kohn–Sham density functional theory (DFT). We verify the accuracy of the formalism by comparing the Hugoniot for carbon with recent Kohn–Sham DFT results in the literature. In so doing, we demonstrate that Kohn–Sham calculations for the Hugoniot can be accelerated by up to two orders of magnitude, while retaining ab initio accuracy. We apply this framework to calculate the Hugoniots of 14 materials in the FPEOS database, comprising 9 single elements and 5 compounds, between temperatures of 10 kK and 2 MK. We find good agreement with first principles results in the literature while providing tighter error bars. In addition, we confirm that the inter-element interaction in compounds decreases with temperature.

70 PLASMA PHYSICS AND FUSION TECHNOLOGY↗

HydraGNN_Predictive_GFM_2024 - Ensemble of predictive graph foundation models for ground state atomistic materials modeling

We provide the ensemble of fifteen pre-trained graph foundation models (GFMs) for atomistic materials modeling applications. Each one of the fifteen GFMs has been trained on five open-source datasets that (once aggregated) amount to over 154 million atomistic structures, which cover over two-thirds of the natural elements of the periodic table and that comprises a broad set of organic and inorganic compounds. This vast set of atomistic structures comprises ground state configurations that are dynamically stable (i.e., equilibrated structures with atomic forces approximately close to zero values) as well as dynamically unstable structures (i.e., non-equilibrium structures with non-negligible non-zero values of atomic forces). The ensemble of datasets aggregated does NOT include excited states. The datasets have been curated to remove atomistic structures with spectral norm of the force tensor above 100 eV/angstrom. Moreover, a linear term of the energy was computed for each dataset using a linear regression model that uses the chemical concentration of each natural element as regressor. The linear term predicted by the linear regression model has been subtracted from each original energy value to perform a re-alignment of the energy values across different electronic structures approximation theories performed to generate the diverse multi-source, multi-fidelity datasets. The folder "ADIOS_files" 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 "ADIOS_files" directory contains 6 sub-directories named as follows: - ANI1x-v3.bp - MPTrj-v3.bp - OC2020-20M-v3.bp - OC2020-v3.bp - OC2022-v3.bp - qm7x-v3.bp Each sub-directory contains the pre-processed datasets converted in Adaptable I/O System (ADIOS) format (https://www.exascaleproject.org/research-project/adios/) that have been used to the development, training, and performance testing of the ensemble go predictive graph foundation models. Each GFM was developed using HydraGNN (https://github.com/ORNL/HydraGNN) as underlying graph neural network (GNN) architecture. The multi-task learning (MTL) capability of HydraGNN was used to simultaneously train the GFMs on labeled values for direct predictions of energy (a total system property of an atomistic structure that measures the chemical stability) and atomic forces (an atomic level property of an atomistic structure that measures the dynamical stability). The hyper parameters of the GFM have been tuned using scalable hyperparameter optimization (HPO) algorithms implemented in the software DeepHyper (https://github.com/deephyper/deephyper). The pre-training of each HPO trial was performed using distributed data parallelism (DDP) to scale the training across 128 compute nodes of the exascale OLCF supercomputer Frontier. Each HPO trial was trained only for 10 epochs and an early stopping was performed to avoid wasting significant computational resources on GNN architectures that were clearly underperforming. For each HPO trial, the 'omnistat' tool developed by (AMD Research - Advanced Micro Device) was used to measure the total energy consumption in kWh. The ensemble of GFMs was obtained by selecting the fifteen best performing HPO trials. Four models have been selected for their clear advantage in accuracy, and these are the GFMs with IDs 229, 156, 147, 260. Additional eleven models have been selected based on judicious balance between accuracy and energy consumption needed for training, and these are the GFMs with IDs 165, 78, 137, 1, 175, 171, 181, 67, 179, 167, 351. Each selected GFM of the ensemble was continued to cumulate a total of at most 30 epochs. In some cases, the total number of epochs actually performed was les than 30 due to two combined factors: (1) the size of the GFM (i.e., the number of model parameters to train) and (2) the total wall-clock time for which the computational resources could be allocated on OLCF-Frontier. The "Ensemble_of_models" directory contains 15 sub-directories named as follows: - gfm_0.229 - gfm_0.156 - gfm_0.147 - gfm_0.260 - gfm_0.165 - gfm_0.78 - gfm_0.137 - gfm_0.1 - gfm_0.175 - gfm_0.171 - gfm_0.181 - gfm_0.67 - gfm_0.179 - gfm_0.167 - gfm_0.351 Each one of these sub-directories refers to one of the fifteen HPO trials that have been selected to continue the pre-training with at most 30 epochs. With each sub-directory associated with a specific HPO trial, the following files can be found: - config.json: file for argument parsing to develop and train an HydraGNN architecture - gfm_0.ID_epoch_N.pk: file with model parameters for HPO ID trial after N epochs of training The ensemble of fifteen GFM architectures was used for (1) ensemble averaging to stabilize the predictions of energy and atomic forces after pre-training for post-processing analysis and (2) ensemble uncertainty quantification (UQ). The code used to develop, pre-train, and load the pre-trained models for post-processing analysis is available on the ORNL-GitHub at the following link: https://github.com/ORNL/HydraGNN/tree/Predictive_GFM_2024

36 MATERIALS SCIENCE↗