A likelihood ratio test for shrinkage covariance estimators
Explore the source record for details and available documents.
Anderson, Dylan Zachary↗
SEARCH · Engineering Papers
Results for “likelihood ratio”
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.
Learning likelihood ratios with neural network classifiers
The likelihood ratio is a crucial quantity for statistical inference in science that enables hypothesis testing, construction of confidence intervals, reweighting of distributions, and more. Many modern scientific applications, however, make use of data- or simulation-driven models for which computing the likelihood ratio can be very difficult or even impossible. By applying the so-called “likelihood ratio trick,” approximations of the likelihood ratio may be computed using clever parametrizations of neural network-based classifiers. A number of different neural network setups can be defined to satisfy this procedure, each with varying performance in approximating the likelihood ratio when using finite training data. We present a series of empirical studies detailing the performance of several common loss functionals and parametrizations of the classifier output in approximating the likelihood ratio of two univariate and multivariate Gaussian distributions as well as simulated high-energy particle physics datasets.
72 PHYSICS OF ELEMENTARY PARTICLES AND FIELDS↗
New Machine Learning Techniques for Simulation-Based Inference: InferoStatic Nets, Kernel Score Estimation, and Kernel Likelihood Ratio Estimation
We propose an intuitive, machine-learning approach to multiparameter inference, dubbed the InferoStatic Networks (ISN) method, to model the score and likelihood ratio estimators in cases when the probability density can be sampled but not computed directly. The ISN uses a backend neural network that models a scalar function called the inferostatic potential $\varphi$. In addition, we introduce new strategies, respectively called Kernel Score Estimation (KSE) and Kernel Likelihood Ratio Estimation (KLRE), to learn the score and the likelihood ratio functions from simulated data. We illustrate the new techniques with some toy examples and compare to existing approaches in the literature. We mention en passant some new loss functions that optimally incorporate latent information from simulations into the training procedure.
72 PHYSICS OF ELEMENTARY PARTICLES AND FIELDS↗
New machine learning techniques for simulation-based inference: InferoStatic nets, kernel score estimation, and kernel likelihood ratio estimation
We propose an intuitive, machine-learning approach to multiparameter inference, dubbed the InferoStatic Networks (ISN) method, to model the score and likelihood ratio estimators in cases when the probability density can be sampled but not computed directly. The ISN uses a backend neural network that models a scalar function called the inferostatic potential \varphi φ . In addition, we introduce new strategies, respectively called Kernel Score Estimation (KSE) and Kernel Likelihood Ratio Estimation (KLRE), to learn the score and the likelihood ratio functions from simulated data. We illustrate the new techniques with some toy examples and compare to existing approaches in the literature. We mention en passant some new loss functions that optimally incorporate latent information from simulations into the training procedure.
Kong, Kyoungchul↗
Inferring subhalo effective density slopes from strong lensing observations with neural likelihood-ratio estimation
ABSTRACT Strong gravitational lensing has emerged as a promising approach for probing dark matter (DM) models on sub-galactic scales. Recent work has proposed the subhalo effective density slope as a more reliable observable than the commonly used subhalo mass function. The subhalo effective density slope is a measurement independent of assumptions about the underlying density profile and can be inferred for individual subhaloes through traditional sampling methods. To go beyond individual subhalo measurements, we leverage recent advances in machine learning and introduce a neural likelihood-ratio estimator to infer an effective density slope for populations of subhaloes. We demonstrate that our method is capable of harnessing the statistical power of multiple subhaloes (within and across multiple images) to distinguish between characteristics of different subhalo populations. The computational efficiency warranted by the neural likelihood-ratio estimator over traditional sampling enables statistical studies of DM perturbers and is particularly useful as we expect an influx of strong lensing systems from upcoming surveys.
Astronomy & Astrophysics↗
Assessing the Effectiveness of Generalized Likelihood Ratio Test Detector Schemes in Seismic Event Detection and the Avoidance of Nontarget Signals
Cross-correlation techniques have played a long-standing and pivotal role in seismic event monitoring. However, the performance of correlation-based detectors is challenged by nuisance seismicity, or nontarget signals. Such detections are a problem when the mission is to automatically map events to the correct source region. Using aftershocks of the 2014 $M_w$ 6.0 South Napa, California, earthquake, we demonstrate the effectiveness of utilizing a dynamic correlation processor framework in a generalized likelihood ratio test (GLRT) detector configuration to minimize nontarget detections. A GLRT maximizes a detection statistic with respect to one or more unknown parameters. In this case, the detection statistic is a template signal match against the waveform in a window sliding over a data stream, and the unknown parameter is an index variable indicating group membership of the template event or events. Detected events are assigned to the event group that yields the largest detection statistic. In this work, our results show that a GLRT detector will outperform a suite of independently operating correlation and subspace detectors in terms of having a lower nontarget detection rate at a given missed detection rate. We also show that a GLRT detector composed of a few high-rank subspace detectors has a slightly higher nontarget detection rate, but a significantly lower missed detection rate, than a GLRT detector composed of many low-rank subspace detectors. The high-rank GLRT configuration produced impressive results even with marginal data (single channel, single station, and very low time bandwidth product), which bodes well for the utility of building efficient aftershock classification systems and global monitoring systems at larger scales. However, future work is required to assess performance at the regional scale and to assess the performance of the system at detecting target events not used in the detector template creation.
58 GEOSCIENCES↗
Bearing Fault Detection on Wind Turbine Gearbox Vibrations Using Generalized Likelihood Ratio-Based Indicators
Studies in condition monitoring literature often aim to detect rolling element bearing faults because they have one of the biggest shares among defects in turbo machinery. Accordingly, several prognosis and diagnosis methods have been devised to identify fault signatures from vibration signals. A recently proposed method to capture the rolling element bearing degradation provides the groundwork for new indicator families utilizing the generalized likelihood ratio test. This novel approach exploits the cyclostationarity and the impulsiveness of vibration signals independently in order to estimate the most suitable indicators for a given fault. However, the method has yet to be tested on complex experimental vibration signals such as those of a wind turbine gearbox. In this study, the approach is applied to the National Renewable Energy Laboratory Wind Turbine Gearbox Condition Monitoring Round Robin Study data set for bearing fault detection purposes. The data set is measured on an experimental test rig of a wind turbine gearbox; hence the complexity of the vibration signals is similar to a real case. The outcome demonstrates that the proposed method is capable of distinguishing between healthy and damaged vibration signals measured on a complex wind turbine gearbox.
condition monitoring↗
Bearing Fault Detection on Wind Turbine Gearbox Vibrations Using Generalized Likelihood Ratio-Based Indicators: Preprint
Studies in condition monitoring literature often aim to detect rolling element bearing faults because they have one of the biggest shares among defects in turbo machinery. Accordingly, several prognosis and diagnosis methods have been devised to identify fault signatures from vibration signals. The underlying idea behind traditional indicators often revolves around tracking both cyclostationarity and abnormal impulses in the vibration signals without distinguishing the two. A recently proposed method to capture the rolling element bearing degradation lays out the groundwork for new indicator families utilizing generalized likelihood ratio test. This novel approach exploits the cyclostationarity and the impulsiveness of vibration signals independently in order to estimate the most suitable indicators for a given fault. However, the method has yet to be tested on complex experimental vibration signals such as those of a wind turbine gearbox. In this study, the approach is applied to the NREL Wind Turbine Gearbox Condition Monitoring Round Robin Study data set for bearing fault detection purposes. The data set is measured on an experimental test rig of a wind turbine gearbox, hence the complexity of the vibration signals is similar to a real case. Furthermore, the new indicators are also tested with signals that carry multiple fault signatures. The outcome demonstrates that the proposed method is capable of distinguishing between healthy and damaged vibration signals measured on a complex wind turbine gearbox.
condition monitoring↗
Codebase release 0.1 for infstat
We propose an intuitive, machine-learning approach to multiparameter inference, dubbed the InferoStatic Networks (ISN) method, to model the score and likelihood ratio estimators in cases when the probability density can be sampled but not computed directly. The ISN uses a backend neural network that models a scalar function called the inferostatic potential \varphi φ . In addition, we introduce new strategies, respectively called Kernel Score Estimation (KSE) and Kernel Likelihood Ratio Estimation (KLRE), to learn the score and the likelihood ratio functions from simulated data. We illustrate the new techniques with some toy examples and compare to existing approaches in the literature. We mention en passant some new loss functions that optimally incorporate latent information from simulations into the training procedure.
Kong, Kyoungchul↗
Discriminative versus generative approaches to simulation-based inference
Most of the fundamental, emergent, and phenomenological parameters of particle and nuclear physics are determined through parametric template fits. Simulations are used to populate histograms which are then matched to data. This approach is inherently lossy, since histograms are binned and low-dimensional. Deep learning has enabled unbinned and high-dimensional parameter estimation through neural likelihood(-ratio) estimation. We compare two approaches for neural simulation-based inference (NSBI): one based on discriminative learning (classification) and one based on generative modeling. These two approaches are directly evaluated on the same datasets, with a similar level of hyperparameter optimization in both cases. In addition to a Gaussian dataset, we study NSBI using a Higgs boson dataset from the FAIR Universe Challenge. We find that both the direct likelihood and likelihood ratio estimation are able to effectively extract parameters with reasonable uncertainties. For the numerical examples and within the set of hyperparameters studied, we found that the likelihood ratio method is more accurate and/or precise. Both methods have a significant spread from the network training and would require ensembling or other mitigation strategies in practice.
high energy physics↗
Optimizers for stabilizing likelihood-free inference
A growing number of applications in particle physics and beyond use neural networks as unbinned likelihood ratio estimators applied to real or simulated data. Precision requirements on the inference tasks demand a high-level of stability from these networks, which are affected by the stochastic nature of training. We show how physics concepts can be used to stabilize network training through a physics-inspired optimizer. In particular, the energy conserving descent (ECD) optimization framework uses classical Hamiltonian dynamics on the space of network parameters to reduce the dependence on the initial conditions while also stabilizing the result near the minimum of the loss function. We develop a version of this optimizer known as , which has few free hyperparameters with limited ranges guided by physical reasoning. We apply to representative likelihood-ratio estimation tasks in particle physics and find on average that it out-performs the widely used Adam optimizer. We expect that ECD will be a useful tool for wide array of data-limited problems, where it is computationally expensive to exhaustively optimize hyperparameters and mitigate fluctuations with ensembling.
Monte Carlo methods↗
Systematic quark/gluon identification with ratios of likelihoods
Discriminating between quark- and gluon-initiated jets has long been a central focus of jet substructure, leading to the introduction of numerous observables and calculations to high perturbative accuracy. At the same time, there have been many attempts to fully exploit the jet radiation pattern using tools from statistics and machine learning. We propose a new approach that combines a deep analytic understanding of jet substructure with the optimality promised by machine learning and statistics. After specifying an approximation to the full emission phase space, we show how to construct the optimal observable for a given classification task. This procedure is demonstrated for the case of quark and gluons jets, where we show how to systematically capture sub-eikonal corrections in the splitting functions, and prove that linear combinations of weighted multiplicity is the optimal observable. In addition to providing a new and powerful framework for systematically improving jet substructure observables, we demonstrate the performance of several quark versus gluon jet tagging observables in parton-level Monte Carlo simulations, and find that they perform at or near the level of a deep neural network classifier. Combined with the rapid recent progress in the development of higher order parton showers, we believe that our approach provides a basis for systematically exploiting subleading effects in jet substructure analyses at the Large Hadron Collider (LHC) and beyond.
72 PHYSICS OF ELEMENTARY PARTICLES AND FIELDS↗
Continuous-time probabilistic models for longitudinal electronic health records
Analysis of longitudinal Electronic Health Record (EHR) data is an important goal for precision medicine. Difficulty in applying Machine Learning (ML) methods, either predictive or unsupervised, stems in part from the heterogeneity and irregular sampling of EHR data. Here, we present an unsupervised probabilistic model that captures nonlinear relationships between variables over continuous-time. This method works with arbitrary sampling patterns and captures the joint probability distribution between variable measurements and the time intervals between them. Inference algorithms are derived that can be used to evaluate the likelihood of future using under a trained model. As an example, we consider data from the United States Veterans Health Administration (VHA) in the areas of diabetes and depression. Likelihood ratio maps are produced showing the likelihood of risk for moderate-severe vs minimal depression as measured by the Patient Health Questionnaire-9 (PHQ-9).
59 BASIC BIOLOGICAL SCIENCES↗
Multivariable degradation modeling and life prediction using multivariate fractional Brownian motion
In system prognostics and health management, multivariable degradation models have been widely developed to predict the life of complex systems using degradation data of multiple Performance Characteristics (PCs). Recent studies have detected a Long-Term Memory (LTM) effect among the degradation process of various PCs, implying a strong coupling phenomenon between the future degradation behavior and historical degradation trajectory. Although the LTM has been widely integrated into single-PC-based degradation modeling, it has not been considered in multi-PC-based scenarios. To capture LTM among multiple PCs, this article proposes a novel LTM-integrated Multivariate Degradation Model (MDM) for system life prediction based on multivariate fractional Brownian motion, which simultaneously incorporates the cross-correlation among different PCs. To estimate parameters of the LTM-integrated MDM, a maximum likelihood method is developed. Here, two likelihood-ratio hypothesis tests are developed to test the existence of the overall and individual LTM effect among multiple PCs. Both simulation studies and physical experiments on the performance degradation of solar energy conversion and storage devices are conducted to validate the proposed model. Results reveal that the proposed LTM-integrated MDM significantly outperforms existing MDMs in life prediction, while the lifetime uncertainty is heavily underestimated by those traditional approaches that neglect the LTM.
42 ENGINEERING↗
An Automated Probabilistic Asteroid Prediscovery Pipeline
We present an automated and probabilistic method to make prediscovery detections of near-Earth asteroids (NEAs) in archival survey images, with the goal of reducing orbital uncertainty immediately after discovery. We refit the Minor Planet Center's astrometry and propagate the full six-parameter covariance to survey epochs to define search regions. We build low-threshold source catalogs for viable images and evaluate every detected source in a search region as a candidate prediscovery. We eliminate false positives by refitting a new orbit to each candidate and probabilistically linking detections across images using a likelihood ratio. Applied to the Zwicky Transient Facility's (ZTF) imaging, we identify approximately 3000 recently discovered NEAs with prediscovery potential, including a doubling of the observational arc for about 500. We use archival ZTF imaging to make prediscovery detections of the potentially hazardous asteroid 2021 DG1, extending its arc by 2.5 yr and reducing future apparition sky plane uncertainty from many degrees to arcseconds. We also recover 2025 FU24 nearly 7 yr before its first known observation, when its sky plane uncertainty covers hundreds of square degrees across thousands of ZTF images. The method is survey agnostic and scalable, enabling rapid orbit refinement for new discoveries from Rubin, NEO Surveyor, and NEOMIR.
79 ASTRONOMY AND ASTROPHYSICS↗
Stochastic agent-based model for predicting turbine-scale raptor movements during updraft-subsidized directional flights
Rapid expansion of wind energy development across the world has highlighted the need to better understand turbine-caused avian mortality. The risk to golden eagles (Aquila chrysaetos) is of particular concern due to their small population size and conservation status. Golden eagles subsidize their flight in part by soaring in orographic updrafts, which can place them in conflict with wind turbines utilizing the same low-altitude wind resource. Understanding the behavior of soaring raptors in varying atmospheric conditions can therefore be relevant to predicting and mitigating their risk of collision. We present a predictive movement model that simulates individual paths of golden eagles during directional flight (such as migration) that is subsidized by orographic updraft. We modeled eagles in a 50 km by 50 km study area in Wyoming containing three wind power plants with documented golden eagle collisions with turbines. The movement model is applicable to any region where ground elevation is known at turbine scale (50 m) and wind conditions are known at facility scale (3 km). For a given set of atmospheric conditions, the model simulates movements of thousands of orographic soaring eagles to produce a density map quantifying the relative probability of eagle presence. We validated the simulated tracks with GPS telemetry data showing four directional tracks made by golden eagles transiting through the area in 2019 and 2020. For each eagle track, validation was performed using the ratio of the model-simulated eagle presence likelihood with uniform eagle presence and the presence computed using directed random-walk movements. We found that the predictive performance of the model was significantly better (likelihood ratio 1) for low-altitude movements than high-altitude movements that can involve thermal-soaring. We employed the model to produce seasonal presence maps for migrating golden eagles. We found significant turbine-level variations in eagle presence between northerly and southerly migration routes through the study area. Overall, the proposed model offers a generalizable, probabilistic, and predictive tool to assist wind energy developers, ecologists, wildlife managers, and industry consultants in estimating the potential for conflict between soaring birds and wind turbines, thereby reducing the need for site-specific data on golden eagle movements.
17 WIND ENERGY↗
Characterizing a supernova’s standing accretion shock instability with neutrinos and gravitational waves
Here, we perform a novel multi-messenger analysis for the identification and parameter estimation of the Standing Accretion Shock Instability (SASI) in a core collapse supernova with neutrino and gravitational wave (GW) signals. In the neutrino channel, this method performs a likelihood ratio test for the presence of SASI in the frequency domain. For gravitational wave signals we process an event with a modified constrained likelihood method. Using simulated supernova signals, the properties of the Hyper-Kamiokande neutrino detector, and O3 LIGO Interferometric data, we produce the two-dimensional probability density distribution (PDF) of the SASI activity indicator and calculate the probability of detection P D as well as the false identification probability P FI . We discuss the probability to establish the presence of the SASI as a function of the source distance in each observational channel, as well as jointly. Compared to a single-messenger approach, the joint analysis results in P D (at P FI = 0.1) of SASI activities that is larger by up to ≈ 40% for a distance to the supernova of 5 kpc. We also discuss how accurately the frequency and duration of the SASI activity can be estimated in each channel separately. Our methodology is suitable for implementation in a realistic data analysis and a multi-messenger setting.
79 ASTRONOMY AND ASTROPHYSICS↗
Statistical Significance Testing for Mixed Priors: A Combined Bayesian and Frequentist Analysis
In many hypothesis testing applications, we have mixed priors, with well-motivated informative priors for some parameters but not for others. The Bayesian methodology uses the Bayes factor and is helpful for the informative priors, as it incorporates Occam’s razor via the multiplicity or trials factor in the look-elsewhere effect. However, if the prior is not known completely, the frequentist hypothesis test via the false-positive rate is a better approach, as it is less sensitive to the prior choice. We argue that when only partial prior information is available, it is best to combine the two methodologies by using the Bayes factor as a test statistic in the frequentist analysis. We show that the standard frequentist maximum likelihood-ratio test statistic corresponds to the Bayes factor with a non-informative Jeffrey’s prior. We also show that mixed priors increase the statistical power in frequentist analyses over the maximum likelihood test statistic. We develop an analytic formalism that does not require expensive simulations and generalize Wilks’ theorem beyond its usual regime of validity. In specific limits, the formalism reproduces existing expressions, such as the p-value of linear models and periodograms. We apply the formalism to an example of exoplanet transits, where multiplicity can be more than 10 7 . We show that our analytic expressions reproduce the $p$-values derived from numerical simulations. We offer an interpretation of our formalism based on the statistical mechanics. We introduce the counting of states in a continuous parameter space using the uncertainty volume as the quantum of the state. We show that both the $p$-value and Bayes factor can be expressed as an energy versus entropy competition.
97 MATHEMATICS AND COMPUTING↗