Results for “Bayesian classification” — Page 8

DOE OSTI2020

Non-Intrusive Appliance Identification with Appliance-Specific Networks

The problem of noninstrusive load monitoring (NILM) is usually formulated as a single-channel blind source separation task, whose successful solution enable fast and convenient load identification and energy disaggregation. When applied at test time, NILM algorithms aim to identify the operating characteristics of individual appliances from an aggregate power measurement of the entire house. Recent advances in deep learning gave rise to many methods that mostly focus on learning a direct mapping from aggregate measurement to individual appliance power. However, these methods are not only computationally expensive, but they often suffer from overfitting and do not generalize very well. In this article, we propose a novel NILM method that leverages advances in statistical learning that have not been properly applied in this domain before. The proposed method consists of three stages: first, a Bayesian nonparametric learning-based approach for appliance state extraction; second, synthetic minority oversampling technique for data augmentation and mitigating the heavy imbalance in switching events; and third, appliance-specific lightweight long short-term memory networks for status classification for each appliance. Here, we adopt a “differential” input (the difference before and after the switching event) to reduce the complexity of network training and make the proposed method robust to multiappliance switching events. Experiments are conducted to demonstrate the effectiveness of the proposed method, achieving superior performance when compared to recent methods. An ablation study is conducted to demonstrate the effectiveness of each module of our method. Finally, we investigate the quality of generated synthetic samples.

42 ENGINEERING↗

DOE OSTI2014

Flexible and scalable methods for quantifying stochastic variability in the era of massive time-domain astronomical data sets

We present the use of continuous-time autoregressive moving average (CARMA) models as a method for estimating the variability features of a light curve, and in particular its power spectral density (PSD). CARMA models fully account for irregular sampling and measurement errors, making them valuable for quantifying variability, forecasting and interpolating light curves, and variability-based classification. We show that the PSD of a CARMA model can be expressed as a sum of Lorentzian functions, which makes them extremely flexible and able to model a broad range of PSDs. We present the likelihood function for light curves sampled from CARMA processes, placing them on a statistically rigorous foundation, and we present a Bayesian method to infer the probability distribution of the PSD given the measured light curve. Because calculation of the likelihood function scales linearly with the number of data points, CARMA modeling scales to current and future massive time-domain data sets. We conclude by applying our CARMA modeling approach to light curves for an X-ray binary, two active galactic nuclei, a long-period variable star, and an RR Lyrae star in order to illustrate their use, applicability, and interpretation.

79 ASTRONOMY AND ASTROPHYSICS↗

DOE OSTI2019

Computer-aided detection using non-convolutional neural network Gaussian processes

Deep convolutional neural networks (CNNs) have in recent years achieved record-breaking performance on many image classification tasks and are therefore well-suited for computer aided detection (CAD). The need for uncertainty quantification for CAD motivates the need for a probabilistic framework for deep learning. The most well-known probabilistic neural network model is the Bayesian neural network (BNN), but BNNs are notoriously difficult to sample for large complex network architectures, and as such their use is restricted to small problems. It is known that the limit of BNNs as their widths increase toward infinity is a Gaussian process (GP), and there has been considerable research interest in these infinitely wide BNNs. Recently, this classic result has been extended to deep architectures in what is termed the neural network Gaussian process (NNGP) model. In this work, we implement an NNGP model and apply it to the ChestXRay14 dataset at the full resolution of 1024x1024 pixels. Even without any convolutional aspects to the network architecture and without any data augmentation, our five layer deep NNGP model outperforms other non-convolutional models and therefore helps to narrow the performance gap between non-convolutional and convolutional models. Our NNGP model is fully Bayesian and therefore offers uncertainty information through its predictive variance that can be used to formulate a predictive confidence measure. We show that the performance of the NNGP model is significantly boosted after low-confidence predictions are rejected, suggesting that convolution is most beneficial only for these low-confidence examples. Finally, our results indicate that an extremely large fully-connected neural network with appropriate regularization could perform as well as the NNGP if not for the computational bottleneck resulting from the large number of model parameters.

Agrawal, Devanshu↗

NASA NTRS1990

Myths and legends in learning classification rules

A discussion is presented of machine learning theory on empirically learning classification rules. Six myths are proposed in the machine learning community that address issues of bias, learning as search, computational learning theory, Occam's razor, universal learning algorithms, and interactive learning. Some of the problems raised are also addressed from a Bayesian perspective. Questions are suggested that machine learning researchers should be addressing both theoretically and experimentally.

Buntine, Wray↗

DOE OSTI2019

Multi-fidelity classification using Gaussian processes: Accelerating the prediction of large-scale computational models

We report that machine learning techniques typically rely on large datasets to create accurate classifiers. However, there are situations when data is scarce and expensive to acquire. This is the case of studies that rely on state-of-the-art computational models which typically take days to run, thus hindering the potential of machine learning tools. In this work, we present a novel classifier that takes advantage of lower fidelity models and inexpensive approximations to predict the binary output of expensive computer simulations. We postulate an autoregressive model between the different levels of fidelity with Gaussian process priors. We adopt a fully Bayesian treatment for the hyper-parameters and use Markov Chain Monte Carlo samplers. We take advantage of the probabilistic nature of the classifier to implement active learning strategies. We also introduce a sparse approximation to enhance the ability of the multi-fidelity classifier to handle a large amount of low fidelity samples. We test these multi-fidelity classifiers against their single-fidelity counterpart with synthetic data, showing a median computational cost reduction of 23% for a target accuracy of 90%. In an application to cardiac electrophysiology, the multi-fidelity classifier achieves an F1 score, the harmonic mean of precision and recall, of 99.6% compared to 74.1% of a single-fidelity classifier when both are trained with 50 samples. In general, our results show that the multi-fidelity classifiers outperform their single-fidelity counterpart in terms of accuracy in all cases. Finally, we envision that this new tool will enable researchers to study classification problems that would otherwise be prohibitively expensive. Source code is available at https://github.com/fsahli/MFclass.

97 MATHEMATICS AND COMPUTING↗

NASA NTRS1990

Myths and legends in learning classification rules

This paper is a discussion of machine learning theory on empirically learning classification rules. The paper proposes six myths in the machine learning community that address issues of bias, learning as search, computational learning theory, Occam's razor, 'universal' learning algorithms, and interactive learnings. Some of the problems raised are also addressed from a Bayesian perspective. The paper concludes by suggesting questions that machine learning researchers should be addressing both theoretically and experimentally.

Buntine, Wray↗

DOE OSTI2026

Optimizing Neutrino Flavor Conversion Measurements through Machine Learning

The phenomenon of neutrino flavor conversion whereby the flavor of a neutrino particle can change between its time of production and later detection was the first definitive evidence of physics beyond the Standard Model. Some of the oscillation parameters used to describe this conversion are not yet well measured, leaving important questions still open regarding flavor conversion both in vacuum and as neutrinos travel through matter. NOvA is a long-baseline neutrino oscillation experiment that uses Fermilab's predominantly $\nu_\mu$ NuMI beam. A 14 kton oil-based liquid scintillator far detector 810 km away is used to measure neutrino oscillation through the $\nu_\mu$ disappearance and $\nu_e$ appearance channels. Super-K is a 50 kton water Cherenkov detector, which measures the disappearance of $\nu_e$ produced during solar fusion. The high density environment of the sun decreases the $\nu_e$ survival probability at higher energies observable in Super-K compared to the vacuum-dominated oscillations at lower energies. However, the transition region is overshadowed by radioactive background in the detector. In both of these experiments, the separation of neutrino detection events from background and classification of neutrino flavor are crucial tasks that benefit from the introduction of machine learning. Chapter 1 gives an overview of neutrinos and the context under which flavor conversion is measured in this dissertation. Chapters 2-7 present the results of a Bayesian sampling approach for the latest NOvA 3-flavor oscillation analysis with $26.6 \times 10^{20}$ protons on target in neutrino mode and $12.5 \times 10^{20}$ in antineutrino mode collected over 10 years. Chapters 8-13 present the results of extending the Super-K solar analysis to lower energies during its fourth phase with 2970 days of livetime.

Yankelevich, Alejandro Jaime [UC, Irvine]↗

DOE OSTI2026

Optimizing Neutrino Flavor Conversion Measurements through Machine Learning

The phenomenon of neutrino flavor conversion whereby the flavor of a neutrino particle can change between its time of production and later detection was the first definitive evidence of physics beyond the Standard Model. Some of the oscillation parameters used to describe this conversion are not yet well measured, leaving important questions still open regarding flavor conversion both in vacuum and as neutrinos travel through matter. NOvA is a long-baseline neutrino oscillation experiment that uses Fermilab's predominantly $\nu_\mu$ NuMI beam. A 14 kton oil-based liquid scintillator far detector 810 km away is used to measure neutrino oscillation through the $\nu_\mu$ disappearance and $\nu_e$ appearance channels. Super-K is a 50 kton water Cherenkov detector, which measures the disappearance of $\nu_e$ produced during solar fusion. The high density environment of the sun decreases the $\nu_e$ survival probability at higher energies observable in Super-K compared to the vacuum-dominated oscillations at lower energies. However, the transition region is overshadowed by radioactive background in the detector. In both of these experiments, the separation of neutrino detection events from background and classification of neutrino flavor are crucial tasks that benefit from the introduction of machine learning. Chapter 1 gives an overview of neutrinos and the context under which flavor conversion is measured in this dissertation. Chapters 2-7 present the results of a Bayesian sampling approach for the latest NOvA 3-flavor oscillation analysis with $26.6 \times 10^{20}$ protons on target in neutrino mode and $12.5 \times 10^{20}$ in antineutrino mode collected over 10 years. Chapters 8-13 present the results of extending the Super-K solar analysis to lower energies during its fourth phase with 2970 days of livetime.

Yankelevich, Alejandro Jaime [UC, Irvine]↗

DOE OSTI2026

Optimizing Neutrino Flavor Conversion Measurements through Machine Learning

The phenomenon of neutrino flavor conversion whereby the flavor of a neutrino particle can change between its time of production and later detection was the first definitive evidence of physics beyond the Standard Model. Some of the oscillation parameters used to describe this conversion are not yet well measured, leaving important questions still open regarding flavor conversion both in vacuum and as neutrinos travel through matter. NOvA is a long-baseline neutrino oscillation experiment that uses Fermilab's predominantly $\nu_\mu$ NuMI beam. A 14 kton oil-based liquid scintillator far detector 810 km away is used to measure neutrino oscillation through the $\nu_\mu$ disappearance and $\nu_e$ appearance channels. Super-K is a 50 kton water Cherenkov detector, which measures the disappearance of $\nu_e$ produced during solar fusion. The high density environment of the sun decreases the $\nu_e$ survival probability at higher energies observable in Super-K compared to the vacuum-dominated oscillations at lower energies. However, the transition region is overshadowed by radioactive background in the detector. In both of these experiments, the separation of neutrino detection events from background and classification of neutrino flavor are crucial tasks that benefit from the introduction of machine learning. Chapter 1 gives an overview of neutrinos and the context under which flavor conversion is measured in this dissertation. Chapters 2-7 present the results of a Bayesian sampling approach for the latest NOvA 3-flavor oscillation analysis with $26.6 \times 10^{20}$ protons on target in neutrino mode and $12.5 \times 10^{20}$ in antineutrino mode collected over 10 years. Chapters 8-13 present the results of extending the Super-K solar analysis to lower energies during its fourth phase with 2970 days of livetime.

Yankelevich, Alejandro Jaime [UC, Irvine] (ORCID:0↗

DOE OSTI2014

The K giant stars from the LAMOST survey data. I. Identification, metallicity, and distance

We present a support vector machine classifier to identify the K giant stars from the LAMOST survey directly using their spectral line features. The completeness of the identification is about 75% for tests based on LAMOST stellar parameters. The contamination in the identified K giant sample is lower than 2.5%. Applying the classification method to about two million LAMOST spectra observed during the pilot survey and the first year survey, we select 298,036 K giant candidates. The metallicities of the sample are also estimated with an uncertainty of 0.13 ∼ 0.29 dex based on the equivalent widths of Mg{sub b} and iron lines. A Bayesian method is then developed to estimate the posterior probability of the distance for the K giant stars, based on the estimated metallicity and 2MASS photometry. The synthetic isochrone-based distance estimates have been calibrated using 7 globular clusters with a wide range of metallicities. The uncertainty of the estimated distance modulus at K = 11 mag, which is the median brightness of the K giant sample, is about 0.6 mag, corresponding to ∼30% in distance. As a scientific verification case, the trailing arm of the Sagittarius stream is clearly identified with the selected K giant sample. Moreover, at about 80 kpc from the Sun, we use our K giant stars to confirm a detection of stream members near the apo-center of the trailing tail. These rediscoveries of the features of the Sagittarius stream illustrate the potential of the LAMOST survey for detecting substructures in the halo of the Milky Way.

79 ASTRONOMY AND ASTROPHYSICS↗

NASA NTRS2004

GLAST Burst Monitor Trigger Classification Algorithm

The Gamma Ray Large Area Space Telescope (GLAST), currently set for launch in the first quarter of 2007, will consist of two instruments, the GLAST Burst Monitor (GBM) and the Large Area Telescope (LAT). One of the goals of the GBM is to identify and locate gamma-ray bursts using on-board software. The GLAST observatory can then be re-oriented to allow observations by the LAT. A Bayesian analysis will be used to distinguish gamma-ray bursts from other triggering events, such as solar flares, magnetospheric particle precipitation, soft gamma repeaters (SGRs), and Cygnus X-1 flaring. The trigger parameters used in the analysis are the burst celestial coordinates, angle from the Earth's horizon, spectral hardness, and the spacecraft geomagnetic latitude. The algorithm will be described and the results of testing will be presented.

Perrin, D. J.↗

NASA NTRS2000

Bayesian Fusion of Color and Texture Segmentations

In many applications one would like to use information from both color and texture features in order to segment an image. We propose a novel technique to combine "soft" segmentations computed for two or more features independently. Our algorithm merges models according to a mean entropy criterion, and allows to choose the appropriate number of classes for the final grouping. This technique also allows to improve the quality of supervised classification based on one feature (e.g. color) by merging information from unsupervised segmentation based on another feature (e.g., texture.)

Manduchi, Roberto↗

DOE OSTI2025

DESI Spectroscopy of HETDEX Emission-line Candidates. I. Line Discrimination Validation

The Hobby–Eberly Dark Energy Experiment (HETDEX) is an untargeted spectroscopic galaxy survey that uses Lyα-emitting galaxies (LAEs) as tracers of 1.9 < z < 3.5 large-scale structure. Most detections consist of a single emission line, whose identity is inferred via a Bayesian analysis of ancillary data. To determine the accuracy of these line identifications, HETDEX detections were observed with the Dark Energy Spectroscopic Instrument (DESI). In two DESI pointings, high-confidence spectroscopic redshifts are obtained for 1157 sources, including 982 LAEs. The DESI spectra are used to evaluate the accuracy of the HETDEX object classifications and tune the methodology to achieve the HETDEX science requirement of ≲2% contamination of the LAE sample by low-redshift emission-line galaxies, while still assigning 96% of the true Lyα emission sample with the correct spectroscopic redshift. We compare emission-line measurements between the two experiments assuming a simple Gaussian line fitting model. Fitted values for the central wavelength of the emission line, the measured line flux, and line widths are consistent between the surveys within uncertainties. Derived spectroscopic redshifts, from the two classification pipelines, when both agree as an LAE classification, are consistent to within $\langle$Δz/(1 + z)$\rangle$ = 6.9 × 10 −5 with an rms scatter of 3.3 × 10 −4 . Data are available at https://data.desi.lbl.gov/desi/public/dr1/vac/dr1/hetdex.

79 ASTRONOMY AND ASTROPHYSICS↗

NASA NTRS1991

Two papers on feed-forward networks

Connectionist feed-forward networks, trained with back-propagation, can be used both for nonlinear regression and for (discrete one-of-C) classification, depending on the form of training. This report contains two papers on feed-forward networks. The papers can be read independently. They are intended for the theoretically-aware practitioner or algorithm-designer; however, they also contain a review and comparison of several learning theories so they provide a perspective for the theoretician. The first paper works through Bayesian methods to complement back-propagation in the training of feed-forward networks. The second paper addresses a problem raised by the first: how to efficiently calculate second derivatives on feed-forward networks.

Buntine, Wray L.↗

DOE OSTI2022

The miniJPAS survey: White dwarf science with 56 optical filters

We analyze the white dwarf population in miniJPAS, the first square degree observed with 56 medium-band, 145 Å in width optical filters by the Javalambre Physics of the accelerating Universe Astrophysical Survey (J-PAS), to provide a data-based forecast for the white dwarf science with low-resolution ($R$ ~ 50) photo-spectra. We define the sample of the bluest point-like sources in miniJPAS with $r$ < 21.5 mag, a point-like probability larger than 0.5, ($u$ - $r$)< 0.80 mag, and ($g$ - $i$)< 0.25 mag. This sample comprises 33 sources with spectroscopic information: 11 white dwarfs and 22 quasi-stellar objects (QSOs). We estimate the effective temperature ($T$ eff ), the surface gravity, and the composition of the white dwarf population by a Bayesian fitting to the observed photo-spectra. The miniJPAS data are sensitive to the Balmer series and the presence of polluting metals. Our results, combined with those from the Javalambre Photometric Local Universe Survey (J-PLUS) which has a lower spectral resolution but has already observed thousands of white dwarfs, suggest that J-PAS photometry would permit – down to $r$ ~ 21.5 mag and at least for sources with 7000 < $T$ eff < 22 000 K – both the classification of the observed white dwarfs into H-dominated and He-dominated with 99% confidence and the detection of calcium absorption for equivalent widths larger than 15 Å. The effective temperature is estimated with a 2% uncertainty, which is close to the 1% from spectroscopy. A precise estimation of the surface gravity depends on the available parallax information. In addition, the white dwarf population at $T$ eff > 7000 K can be segregated from the bluest extragalactic QSOs, providing a clean sample based on optical photometry alone. The J-PAS low-resolution photo-spectra would produce precise effective temperatures and atmospheric compositions for white dwarfs, complementing the data from Gaia. J-PAS will also detect and characterize new white dwarfs beyond the Gaia magnitude limit, providing faint candidates for spectroscopic follow-up.

79 ASTRONOMY AND ASTROPHYSICS↗

NASA NTRS2012

The Analysis of the Contribution of Human Factors to the In-Flight Loss of Control Accidents

In-flight loss of control (LOC) is currently the leading cause of fatal accidents based on various commercial aircraft accident statistics. As the Next Generation Air Transportation System (NextGen) emerges, new contributing factors leading to LOC are anticipated. The NASA Aviation Safety Program (AvSP), along with other aviation agencies and communities are actively developing safety products to mitigate the LOC risk. This paper discusses the approach used to construct a generic integrated LOC accident framework (LOCAF) model based on a detailed review of LOC accidents over the past two decades. The LOCAF model is comprised of causal factors from the domain of human factors, aircraft system component failures, and atmospheric environment. The multiple interdependent causal factors are expressed in an Object-Oriented Bayesian belief network. In addition to predicting the likelihood of LOC accident occurrence, the system-level integrated LOCAF model is able to evaluate the impact of new safety technology products developed in AvSP. This provides valuable information to decision makers in strategizing NASA's aviation safety technology portfolio. The focus of this paper is on the analysis of human causal factors in the model, including the contributions from flight crew and maintenance workers. The Human Factors Analysis and Classification System (HFACS) taxonomy was used to develop human related causal factors. The preliminary results from the baseline LOCAF model are also presented.

Ancel, Ersin↗

NASA NTRSDate not supplied

Infusing Statistical Thinking into the NASA Quesst Community Test Campaign

Statistical thinking permeates many important decisions as NASA plans its Quesst mission, which will culminate in a series of community overflights using the X-59 aircraft to demonstrate low-noise supersonic flight. Month-long longitudinal surveys will be deployed to assess human perception and annoyance to this new acoustic phenomenon. NASA works with a large contractor team to develop systems and methodologies to estimate noise doses, to test and field socio-acoustic surveys, and to study the relationship between the two quantities, dose and response, through appropriate choices of statistical models. This latter dose-response relationship will serve as an important tool as national and international noise regulators debate whether overland supersonic flights could be permitted once again within permissible noise limits. In this presentation we highlight several areas where statistical thinking has come into play, including issues of sampling, classification and data fusion, and analysis of longitudinal survey data that are subject to rare events and the consequences of measurement error. We note several operational constraints that shape the appeal or feasibility of some decisions on statistical approaches, and we identify several important remaining questions to be addressed.

Bayesian model↗

DOE OSTI2023

Chaconne: A Statistical Approach to Nonlocal Compression for Supervised Learning, Semi-Supervised Learning, and Anomaly Detection

This project developed a novel statistical understanding of compression analytics (CA), which has challenged and clarified some core assumptions about CA, and enabled the development of novel techniques that address vital challenges of national security. Specifically, this project has yielded the development of novel capabilities including 1. Principled metrics for model selection in CA, 2. Techniques for deriving/applying optimal classification rules and decision theory to supervised CA, including how to properly handle class imbalance and differing costs of misclassification, 3. Two techniques for handling nonlocal information in CA, 4. A novel technique for unsupervised CA that is agnostic with regard to the underlying compression algorithm, 5. A framework for semisupervised CA when a small number of labels are known in an otherwise large unlabeled dataset. 6. The academic alliance component of this project has focused on the development of a novel exemplar-based Bayesian technique for estimating variable length Markov models (closely related to PPM [prediction by partial matching] compression techniques). We have developed examples illustrating the application of our work to text, video, genetic sequences, and unstructured cybersecurity log files.

99 GENERAL AND MISCELLANEOUS↗