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 91 records · Page 5

Harmonizing tau positron emission tomography in Alzheimer's disease: The CenTauR scale and the joint propagation model

Abstract INTRODUCTION Tau‐positron emission tomography (PET) outcome data of patients with Alzheimer's disease (AD) cannot currently be meaningfully compared or combined when different tracers are used due to differences in tracer properties, instrumentation, and methods of analysis. METHODS Using head‐to‐head data from five cohorts with tau PET radiotracers designed to target tau deposition in AD, we tested a joint propagation model (JPM) to harmonize quantification (units termed “CenTauR” [CTR]). JPM is a statistical model that simultaneously models the relationships between head‐to‐head and anchor point data. JPM was compared to a linear regression approach analogous to the one used in the amyloid PET Centiloid scale. RESULTS A strong linear relationship was observed between CTR values across brain regions. Using the JPM approach, CTR estimates were similar to, but more accurate than, those derived using the linear regression approach. DISCUSSION Preliminary findings using the JPM support the development and adoption of a universal scale for tau‐PET quantification. Highlights Tested a novel joint propagation model (JPM) to harmonize quantification of tau PET. Units of common scale are termed “CenTauRs”. Tested a Centiloid‐like linear regression approach. Using five cohorts with head‐to‐head tau PET, JPM outperformed linearregressionbased approach. Strong linear relationship was observed between CenTauRs values across brain regions.

Neurosciences & Neurology↗

Quantitative ultrasound characterization of locally advanced breast cancer by estimation of its scatterer properties

Purpose: Tumor grading is an important part of breast cancer diagnosis and currently requires biopsy as its standard. Here, the authors investigate quantitative ultrasound parameters in locally advanced breast cancers that can potentially separate tumors from normal breast tissue and differentiate tumor grades. Methods: Ultrasound images and radiofrequency data from 42 locally advanced breast cancer patients were acquired and analyzed. Parameters related to the linear regression of the power spectrum—midband fit, slope, and 0-MHz-intercept—were determined from breast tumors and normal breast tissues. Mean scatterer spacing was estimated from the spectral autocorrelation, and the effective scatterer diameter and effective acoustic concentration were estimated from the Gaussian form factor. Parametric maps of each quantitative ultrasound parameter were constructed from the gated radiofrequency segments in tumor and normal tissue regions of interest. In addition to the mean values of the parametric maps, higher order statistical features, computed from gray-level co-occurrence matrices were also determined and used for characterization. Finally, linear and quadratic discriminant analyses were performed using combinations of quantitative ultrasound parameters to classify breast tissues. Results: Quantitative ultrasound parameters were found to be statistically different between tumor and normal tissue (p < 0.05). The combination of effective acoustic concentration and mean scatterer spacing could separate tumor from normal tissue with 82% accuracy, while the addition of effective scatterer diameter to the combination did not provide significant improvement (83% accuracy). Furthermore, the two advanced parameters, including effective scatterer diameter and mean scatterer spacing, were found to be statistically differentiating among grade I, II, and III tumors (p = 0.014 for scatterer spacing, p = 0.035 for effective scatterer diameter). The separation of the tumor grades further improved when the textural features of the effective scatterer diameter parametric map were combined with the mean value of the map (p = 0.004). Conclusions: Overall, the binary classification results (tumor versus normal tissue) were more promising than tumor grade assessment. Combinations of advanced parameters can further improve the separation of tumors from normal tissue compared to the use of linear regression parameters. While the linear regression parameters were sufficient for characterizing breast tumors and normal breast tissues, advanced parameters and their textural features were required to better characterize tumor subtypes.

62 RADIOLOGY AND NUCLEAR MEDICINE↗

Feature extraction of event-related potentials using wavelets: an application to human performance monitoring

This report describes the development and evaluation of mathematical models for predicting human performance from discrete wavelet transforms (DWT) of event-related potentials (ERP) elicited by task-relevant stimuli. The DWT was compared to principal components analysis (PCA) for representation of ERPs in linear regression and neural network models developed to predict a composite measure of human signal detection performance. Linear regression models based on coefficients of the decimated DWT predicted signal detection performance with half as many free parameters as comparable models based on PCA scores. In addition, the DWT-based models were more resistant to model degradation due to over-fitting than PCA-based models. Feed-forward neural networks were trained using the backpropagation algorithm to predict signal detection performance based on raw ERPs, PCA scores, or high-power coefficients of the DWT. Neural networks based on high-power DWT coefficients trained with fewer iterations, generalized to new data better, and were more resistant to overfitting than networks based on raw ERPs. Networks based on PCA scores did not generalize to new data as well as either the DWT network or the raw ERP network. The results show that wavelet expansions represent the ERP efficiently and extract behaviorally important features for use in linear regression or neural network models of human performance. The efficiency of the DWT is discussed in terms of its decorrelation and energy compaction properties. In addition, the DWT models provided evidence that a pattern of low-frequency activity (1 to 3.5 Hz) occurring at specific times and scalp locations is a reliable correlate of human signal detection performance. Copyright 1999 Academic Press.

Evoked Potentials↗

Feature Extraction of Event-Related Potentials Using Wavelets: An Application to Human Performance Monitoring

This report describes the development and evaluation of mathematical models for predicting human performance from discrete wavelet transforms (DWT) of event-related potentials (ERP) elicited by task-relevant stimuli. The DWT was compared to principal components analysis (PCA) for representation of ERPs in linear regression and neural network models developed to predict a composite measure of human signal detection performance. Linear regression models based on coefficients of the decimated DWT predicted signal detection performance with half as many f ree parameters as comparable models based on PCA scores. In addition, the DWT-based models were more resistant to model degradation due to over-fitting than PCA-based models. Feed-forward neural networks were trained using the backpropagation,-, algorithm to predict signal detection performance based on raw ERPs, PCA scores, or high-power coefficients of the DWT. Neural networks based on high-power DWT coefficients trained with fewer iterations, generalized to new data better, and were more resistant to overfitting than networks based on raw ERPs. Networks based on PCA scores did not generalize to new data as well as either the DWT network or the raw ERP network. The results show that wavelet expansions represent the ERP efficiently and extract behaviorally important features for use in linear regression or neural network models of human performance. The efficiency of the DWT is discussed in terms of its decorrelation and energy compaction properties. In addition, the DWT models provided evidence that a pattern of low-frequency activity (1 to 3.5 Hz) occurring at specific times and scalp locations is a reliable correlate of human signal detection performance.

Trejo, Leonard J.↗

Determination of elemental impurities in plastic calibration standards using laser induced breakdown spectroscopy

Dual-energy computed tomography (CT) scanning is a rapidly emerging imaging technique employed in nondestructive evaluation of various materials. CT has been used for characterizing rocks and visualizing multiphase flow through rocks for over 25 years. The most common technique for dual-energy CT scanning relies on homogeneous calibration standards to produce the most accurate decoupled data. However, the use of calibration standards with impurities increases the probability of error in the reconstructed data and results in poor rock characterization. Laser-induced breakdown spectroscopy was used to determine impurity concentration in a set of commercially purchased calibration standards used in dual-energy scanning for material identification with coal samples. Two calibration models were developed by using univariate calibration with the internal ratio method and multiple linear regression. Seven elements (Al, Fe, Mg, Na, Ni, Sr, and Ti) were examined in five different samples containing varying amounts of each ion to compare calibration from univariate data analysis and from multivariate data analysis. The contaminant concentrations were also measured by a commercially available inductively coupled plasma optical emission spectroscopy instrument, and the data were used as a reference in developing calibration curves for a modified version of the single linear regression model and the multiple linear regression model.

01 COAL, LIGNITE, AND PEAT↗

Lowering post‐construction yield assessment uncertainty through better wind plant power curves

Abstract Many operational analyses of wind power plants require a statistical relationship, which can be called the wind plant power curve, to be developed between wind plant energy production and concurrent atmospheric variables. Currently, a univariate linear regression at monthly resolution is the industry standard for post‐construction yield assessments. Here, we evaluate the benefits in augmenting this conventional approach by testing alternative regressions performed with multiple inputs, at a finer time resolution, and using nonlinear machine‐learning algorithms. We utilize the National Renewable Energy Laboratory's open‐source software package OpenOA to assess wind plant power curves for 10 wind plants. When a univariate generalized additive model at daily or hourly resolution is used, regression uncertainty is reduced, in absolute terms, by up to 1.0 % and 1.2 % (corresponding to a −59 % and −80 % relative change), respectively, compared to a univariate linear regression at monthly resolution; also, a more accurate assessment of the mean long‐term wind plant production is achieved. Additional input variables also reduce the regression uncertainty: when temperature is added as an input to the conventional monthly linear regression, the operational analysis uncertainty connected to regression is reduced, in absolute terms, by up to 0.5 % (−43 % relative change) for wind power plants with strong seasonal variability. Adding input variables to the machine‐learning model at daily resolution can further reduce regression uncertainty, with up to a −10 % relative change. Based on these results, we conclude that a multivariate nonlinear regression at daily or hourly resolution should be recommended for assessing wind plant power curves.

17 WIND ENERGY↗

Cyst-based measurements for assessing lymphangioleiomyomatosis in computed tomography

Purpose: To investigate the efficacy of a new family of measurements made on individual pulmonary cysts extracted from computed tomography (CT) for assessing the severity of lymphangioleiomyomatosis (LAM). Methods: CT images were analyzed using thresholding to identify a cystic region of interest from chest CT of LAM patients. Individual cysts were then extracted from the cystic region by the watershed algorithm, which separates individual cysts based on subtle edges within the cystic regions. A family of measurements were then computed, which quantify the amount, distribution, and boundary appearance of the cysts. Sequential floating feature selection was used to select a small subset of features for quantification of the severity of LAM. Adjusted R{sup 2} from multiple linear regression and R{sup 2} from linear regression against measurements from spirometry were used to compare the performance of our proposed measurements with currently used density based CT measurements in the literature, namely, the relative area measure and the D measure. Results: Volumetric CT data, performed at total lung capacity and residual volume, from a total of 49 subjects enrolled in the MILES trial were used in our study. Our proposed measures had adjusted R{sup 2} ranging from 0.42 to 0.59 when regressing against the spirometry measures, with p < 0.05. For previously used density based CT measurements in the literature, the best R{sup 2} was 0.46 (for only one instance), with the majority being lower than 0.3 or p > 0.05. Conclusions: The proposed family of CT-based cyst measurements have better correlation with spirometric measures than previously used density based CT measurements. They show potential as a sensitive tool for quantitatively assessing the severity of LAM.

60 APPLIED LIFE SCIENCES↗

Modeling Solar Lyman Alpha Irradiance

Solar Lyman alpha irradiance is estimated from various solar indices using linear regression analyses. Models developed with multiple linear regression analysis, including daily values and 81-day running means of solar indices, predict reasonably well both the short- and long-term variations observed in Lyman alpha. It is shown that the full disk equivalent width of the He line at 1083 nm offers the best proxy for Lyman alpha, and that the total irradiance corrected for sunspot effect also has a high correlation with Lyman alpha.

Pap, J.↗

Effect of Malmquist bias on correlation studies with IRAS data base

The relationships between galaxy properties in the sample of Trinchieri et al. (1989) are reexamined with corrections for Malmquist bias. The linear correlations are tested and linear regressions are fit for log-log plots of L(FIR), L(H-alpha), and L(B) as well as ratios of these quantities. The linear correlations for Malmquist bias are corrected using the method of Verter (1988), in which each galaxy observation is weighted by the inverse of its sampling volume. The linear regressions are corrected for Malmquist bias by a new method invented here in which each galaxy observation is weighted by its sampling volume. The results of correlation and regressions among the sample are significantly changed in the anticipated sense that the corrected correlation confidences are lower and the corrected slopes of the linear regressions are lower. The elimination of Malmquist bias eliminates the nonlinear rise in luminosity that has caused some authors to hypothesize additional components of FIR emission.

Verter, Frances↗

Multi-Variate LSTM Prediction of Alaska Magnetometer Chain Utilizing a Coupled Model Approach

During periods of rapidly changing geomagnetic conditions electric fields form within the Earth’s surface and induce currents known as geomagnetically induced currents(GICs), which interact with unprotected electrical systems our society relies on. In this study, we train multi-variate Long-Short Term Memory neural networks to predict magnitude of north-south component of the geomagnetic field (|BN|) at multiple ground magnetometer stations across Alaska provided by the SuperMAG database with a future goal of predicting geomagnetic field disturbances. Each neural network is driven by solar wind and interplanetary magnetic field inputs from the NASA OMNI database spanning from 2000–2015 and is fine tuned for each station to maximize the effectiveness in predicting |BN|. The neural networks are then compared against multivariate linear regression models driven with the same inputs at each station using Heidke skill scores with thresholds at the 50, 75, 85, and 99 percentiles for |BN|. The neural network models show significant increases over the linear regression models for |BN| thresholds. We also calculate the Heidke skill scores for d|BN|/dt by deriving d|BN|/dt from |BN| predictions. However, neural network models do not show clear outperformance compared to the linear regression models. To retain the sign information and thus predict BN instead of |BN|, a secondary so-called polarity model is utilized. The polarity model is run in tandem with the neural networks predicting geomagnetic field in a coupled model approach and results in a high correlation between predicted and observed values for all stations. We find this model a promising starting point for a machine learned geomagnetic field model to be expanded upon through increased output time history and fast turnaround times.

Matthew Blandin↗

Evaluating the potential of short-term instrument deployment to improve distributed wind resource assessment

Distributed wind projects, which are connected at the distribution level of an electricity system or in off-grid applications to serve specific or local energy needs, often rely solely on wind resource models to establish wind speed and energy generation expectations. Historically, anemometer loan programs have provided an affordable avenue for more accurate onsite wind resource assessment, and the lowering cost of lidar systems has shown similar advantages for more recent assessments. While a full 12 months of onsite wind measurement is the standard for correcting model-based long-term wind speed estimates for utility-scale wind farms, the time and capital investment involved in gathering onsite measurements must be reconciled with the energy needs and funding opportunities that drive expedient deployment of distributed wind projects. Much literature exists to quantify the performance of correcting long-term wind speed estimates with 1 or more years of observational data, but few studies explore the impacts of correcting with months-long observational periods. This study aims to answer the question of how short you can go in terms of the observational time period needed to make impactful improvements to model-based long-term wind speed estimates. Three algorithms, multivariable linear regression, adaptive regression splines, and regression trees, are evaluated for their skill at correcting long-term wind resource estimates from the European Centre for Medium-Range Weather Forecasts Reanalysis version 5 (ERA5) using months-long periods of observational data from 66 locations across the US. On average, correction with even 1 month of observations provides significant improvement over the baseline ERA5 wind speed estimates and produces median bias magnitudes and relative errors within 0.22 m s −1 and 4 percentage points of the median bias magnitudes and relative errors achieved using the standard 12 months of data for correction. However, in cases when the shortest observational periods (1 to 2 months) used for correction are not well correlated with the overlapping ERA5 reference, the resultant long-term wind speed errors are worse than those produced using ERA5 without correction. Summer months, which are characterized by weaker relative wind speeds and standard deviations for most of the evaluation sites, tend to produce the worst results for long-term correction using months-long observations. The three tested algorithms perform similarly for long-term wind speed bias; however, regression trees perform notably worse than multivariable linear regression and adaptive regression splines in terms of correlation when using 6 months or less of observational data for correction. Translating the analysis to wind energy, median relative errors in the capacity factor are on average within 10 % using 1 month of training. If the observation period used for correction is not well correlated with the reference data, however, misrepresentation of the observed capacity factor can be substantial. The risk associated with poor correlation between the observed and reference datasets decreases with increasing training period length. In the worst-correlation scenarios, the median capacity factor relative errors from using 1, 3, and 6 months are within 47 %, 26 %, and 16 %, respectively.

17 WIND ENERGY↗

Associations of blood lead, cadmium, and mercury with estimated glomerular filtration rate in the Korean general population: Analysis of 2008-2010 Korean National Health and Nutrition Examination Survey data

Introduction: The objective of this study was to evaluate associations between blood lead, cadmium, and mercury levels with estimated glomerular filtration rate in a general population of South Korean adults. Methods: This was a cross-sectional study based on data obtained in the Korean National Health and Nutrition Examination Survey (KNHANES) (2008-2010). The final analytical sample consisted of 5924 participants. Estimated glomerular filtration rate (eGFR) was calculated using the MDRD Study equation as an indicator of glomerular function. Results: In multiple linear regression analysis of log2-transformed blood lead as a continuous variable on eGFR, after adjusting for covariates including cadmium and mercury, the difference in eGFR levels associated with doubling of blood lead were -2.624 mL/min per 1.73 m Superscript-Two (95% CI: -3.803 to -1.445). In multiple linear regression analysis using quartiles of blood lead as the independent variable, the difference in eGFR levels comparing participants in the highest versus the lowest quartiles of blood lead was -3.835 mL/min per 1.73 m Superscript-Two (95% CI: -5.730 to -1.939). In a multiple linear regression analysis using blood cadmium and mercury, as continuous or categorical variables, as independent variables, neither metal was a significant predictor of eGFR. Odds ratios (ORs) and 95% CI values for reduced eGFR calculated for log2-transformed blood metals and quartiles of the three metals showed similar trends after adjustment for covariates. Discussion: In this large, representative sample of South Korean adults, elevated blood lead level was consistently associated with lower eGFR levels and with the prevalence of reduced eGFR even in blood lead levels below 10 {mu}g/dL. In conclusion, elevated blood lead level was associated with lower eGFR in a Korean general population, supporting the role of lead as a risk factor for chronic kidney disease.

60 APPLIED LIFE SCIENCES↗

EBSD seed LDRD project: Does Corona Virus – 2019 (COVID-19) and Seasonal Flu have similar meteorology and air quality controls driving their spread?

Seasonal influenza and Influenza like Illnesses (ILI) pose a serious public health risk and in turn affect the economy. Various factors affect ILI cases and mortality, including the pathogen and its interaction with the host, as well as environmental and socioeconomic factors such as meteorology, household structure, air pollution, urbanization, and population. Despite the growing number of studies on influenza and ILI, challenges remain in forecasting the timing of seasonal onset, outbreak patterns, and key factors affecting transmission. In particular, the impacts of meteorology and air quality on ILI have been challenging to understand, with linear regression studies focused on different geographic regions producing contradictory results. For example, influenza seasonality has been associated with cold-dry conditions in temperate mid-latitudes, but with humid-rainy conditions in tropical climates. These apparently contradictory results imply that the relationships between influenza cases and atmospheric variables may be too complex to be captured by linear regression models. In this seed project, we analyze meteorology and air quality variables from numerical models to determine which atmospheric variables are most helpful in predicting weekly changes in recorded flu cases. In contrast to most previous studies that relied on linear regression analysis to predict the timing of the flu onset or peak, we employ a robust machine learning algorithm to evaluate the contribution of atmospheric variables to weekly changes in recorded ILI cases. These results may also be relevant to the spread of other respiratory illnesses such as Corona Virus Disease – 2019 (COVID-19).

60 APPLIED LIFE SCIENCES↗

Persistent organic pollutants, skull size and bone density of polar bears (Ursus maritimus) from East Greenland 1892–2015 and Svalbard 1964–2004

Highlights: • Density in East Greenland males decreased over time. • Size in East Greenland females tented to decrease over time. • Size was highest in Svalbard polar bears. • Density was positive correlated to POPs. • Environmental changes and POPs may affect bone density and size. We investigated skull size (condylobasal length; CBL) and bone mineral density (BMD) in polar bears (Ursus maritimus) from East Greenland (n = 307) and Svalbard (n = 173) sampled during the period 1892–2015 in East Greenland and 1964–2004 at Svalbard. Adult males from East Greenland showed a continuous decrease in BMD from 1892 to 2015 (linear regression: p < 0.01) indicating that adult male skulls collected in the early pre-pollution period had the highest BMD. A similar decrease in BMD over time was not found for the East Greenland adult females. However, there was a non-significant trend that the skull size of adult East Greenland females was negatively correlated with collection year 1892–2015 (linear regression: p = 0.06). No temporal change was found for BMD or skull size in Svalbard polar bears (ANOVA: all p > 0.05) nor was there any significant difference in BMD between Svalbard and East Greenland subpopulations. Skull size was larger in polar bears from Svalbard than from East Greenland (two-way ANOVA: p = 0.003). T-scores reflecting risk of osteoporosis showed that adult males from both East Greenland and Svalbard are at risk of developing osteopenia. Finally, when correcting for age and sex, BMD in East Greenland polar bears increased with increasing concentrations of persistent organic pollutants (POPs) i.e. ΣPCB (polychlorinated biphenyls), ΣHCH (hexachlorohexane), HCB (hexachlorobenzene) and ΣPBDE (polybrominated diphenyl ethers) while skull size increased with ΣHCH concentrations all in the period 1999–2014 (multiple linear regression: all p < 0.05, n = 175). The results suggest that environmental changes over time, including exposure to POPs, may affect bone density and size of polar bears.

54 ENVIRONMENTAL SCIENCES↗

A New Methodology for Calculating The Energy Performance of Manufacturing Facilities

Total energy comparison (TEC), classic energy intensity (CEI), and linear regression models (LRM) are the three most common approaches used to track the energy performance of manufacturing facilities. TEC simply compares the total energy consumption from utility bills. TEC is rarely used because it does not consider the variation of any factors that may affect energy consumption. One step better, CEI considers the variation of production rates by using the ratio of annual total energy consumption over annual total production. However, CEI fundamentally assumes that energy consumption is zero if the production rate is zero. This is almost never true and can cause significant errors. Using linear regression models, LRM considers the impact of multiple variables and, therefore, most accurately tracks energy performance. Unfortunately, because of the lack of either required data or technical expertise, some facilities cannot create valid linear regression models. For these cases, CEI is the only option. Here, the goal of this article is to develop a new methodology, modified energy intensity (MEI), which is more accurate than CEI, but requires much less data and is easier to implement than LRM. Using the underlying principle of LRM, this article first outlines the mathematical derivation of the equations for MEI, then explains them from an engineering perspective. Finally, the implementation of MEI is discussed.

32 ENERGY CONSERVATION, CONSUMPTION, AND UTILIZATI↗

Composition-Based Density Model for High Level Waste Glasses

In this report, the Savannah River National Laboratory (SRNL) provides a first-principles model capable of predicting the density of high-level waste (HLW) glass based on the glass composition. The model relies on the additivity of the specific volume of bound glass oxides to obtain a quantitative evaluation of the approximate glass density. A database of glass densities and compositions was compiled from various sources including previous Defense Waste Processing Facility (DWPF) variability study glasses, HLW studies from other institutions, and literature references. There are a total of 1104 glass compositions in the database with densities ranging from 2.224 g/cm 3 to 3.997 g/cm 3 . The breadth of the compositional coverage space is reported as well. A predictive method was investigated in which the glass density was calculated as a function of the weighted partial molar volumes of the individual glass oxides and the molar mass of the glass. This method yielded a high correlation between the predicted density and the measured density. In addition, a linear regression of these predictions indicated that the predicted value was equal to the measured within the bounds of analytical uncertainty. The following conclusions were drawn from the results of these analyses: 1. Measured glass density is highly correlated with glass composition (R 2 ≈ 0.95). 2. The correlation between glass density and glass composition can be expressed as either: a. A weighted sum of the individual oxide glass-component densities, or b. A weighted sum of the individual oxide glass-component partial molar volumes. 3. A linear regression of the predicted density via partial molar volume method to the measured density yields a slope of 0.9975 and an intercept of -0.0007 – the uncertainty interval associated with these coefficients contain one and zero, respectively. 4. The fitted coefficients and associated uncertainty intervals of the linear regression of the predicted density via the partial molar volume method to the measured density suggest that the predicted theoretical density is, within analysis uncertainty, equal to the measured density of the glass. It is recommended that glass density measurements included with the variability study for each sludge batch be discontinued. In their place, SRNL will calculate a bounding glass density value for the frit/sludge system using the model. It is also recommended that DWPF implement the density model as part of the fissile loading determination portion of the SME acceptability process. Details for implementation are provided in this report.

12 MANAGEMENT OF RADIOACTIVE AND NON-RADIOACTIVE W↗

Using Ancillary Information to Improve Hypocenter Estimation: Bayesian Single Event Location (BSEL).

We have developed and tested an algorithm, Bayesian Single Event Location (BSEL), for esti- mating the location of a seismic event. The estimation approach di*ers from established non-linear regression techniques by using a Bayesian prior probability density function (prior PDF) to incor- porate ancillary physical basis constraints about event location. P wave arrival times from seismic events are used in the development. Depth, a focus of this paper, may be modeled with a prior PDF (potentially skewed) that captures physical basis bounds from ancillary event characteristics. For instance the surface wave Rg is present in a waveform only when an event is shallow. A high- condence Rg detection in one or more event waveforms implies a shallow-skewed prior PDF for the depth parameter. This PDF is constructed from a physically-based Rayleigh wave depth excitation eigenfunction that is based on the minimum period of observation from a spectrogram analysis and estimated near-source elastic parameters. The proposed Bayesian algorithm is illustrated with events that demonstrate its congruity with established hypocenter estimation methods and its application potential. The BSEL method is applied to the Mw 5.6 Dillon, MT earthquake of July 26, 2005 and a shallow Mw 4 earthquake that occurred near Bardwell, KY on June 6, 2003. In both cases we simulate BSEL on a small subset of arrival time data to illustrate the power of the technique. No Rg was actually observed for the Dillon, MT earthquake, but we used the minimum observed period of a Rayleigh wave (7 seconds) to reduce the depth and origin time uncertainty. A strong Rg was observed from the Bardwell, KY earthquake that places very strong constraints on depth and origin time.

58 GEOSCIENCES↗

Leveraging Multitime Hamilton–Jacobi PDEs for Certain Scientific Machine Learning Problems

Hamilton-Jacobi partial differential equations (HJ PDEs) have deep connections with a wide range of fields, including optimal control, differential games, and imaging sciences. By considering the time variable to be a higher dimensional quantity, HJ PDEs can be extended to the multi-time case. In this paper, we establish a novel theoretical connection between specific optimization problems arising in machine learning and the multi-time Hopf formula, which corresponds to a representation of the solution to certain multi-time HJ PDEs. Through this connection, we increase the interpretability of the training process of certain machine learning applications by showing that when we solve these learning problems, we also solve a multi-time HJ PDE and, by extension, its corresponding optimal control problem. As a first exploration of this connection, we develop the relation between the regularized linear regression problem and the Linear Quadratic Regulator (LQR). We then leverage our theoretical connection to adapt standard LQR solvers (namely, those based on the Riccati ordinary differential equations) to design new training approaches for machine learning. Lastly, we provide some numerical examples that demonstrate the versatility and possible computational advantages of our Riccati-based approach in the context of continual learning, post-training calibration, transfer learning, and sparse dynamics identification.

97 MATHEMATICS AND COMPUTING↗