Search NASA⌕ Search

SEARCH · Search NASA

Results for “maximum likelihood estimation”

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 937 records · Page 52

A Conceptual Approach to Assimilating Remote Sensing Data to Improve Soil Moisture Profile Estimates in a Surface Flux/Hydrology Model: Overview - Part 1

Knowledge of the amount of water in the soil is of great importance to many earth science disciplines. Soil moisture is a key variable in controlling the exchange of water and energy between the land surface and the atmosphere. Thus, soil moisture information is valuable in a wide range of applications including weather and climate, runoff potential and flood control, early warning of droughts, irrigation, crop yield forecasting, soil erosion, reservoir management, geotechnical engineering, and water quality. Despite the importance of soil moisture information, widespread and continuous measurements of soil moisture are not possible today. Although many earth surface conditions can be measured from satellites, we still cannot adequately measure soil moisture from space. Research in soil moisture remote sensing began in the mid 1970s shortly after the surge in satellite development. Recent advances in remote sensing have shown that soil moisture can be measured, at least qualitatively, by several methods. Quantitative measurements of moisture in the soil surface layer have been most successful using both passive and active microwave remote sensing, although complications arise from surface roughness and vegetation type and density. Early attempts to measure soil moisture from space-borne microwave instruments were hindered by what is now considered sub-optimal wavelengths (shorter than 5 cm) and the coarse spatial resolution of the measurements. L-band frequencies between 1 and 3 GHz (10-30 cm) have been deemed optimal for detection of soil moisture in the upper few centimeters of soil. The Electronically Steered Thinned Array Radiometer (ESTAR), an aircraft-based instrument operating a 1,4 GHz, has shown great promise for soil moisture determination. Initiatives are underway to develop a similar instrument for space. Existing space-borne synthetic aperture radars (SARS) operating at C- and L-band have also shown some potential to detect surface wetness. The advantage of radar is its much higher resolution than passive microwave systems, but it is currently hampered by surface roughness effects and the lack of a good algorithm based on a single frequency and single polarization. In addition, its repeat frequency is generally low (about 40 days). In the meantime, two new radiometers offer some hope for remote sensing of soil moisture from space. The Tropical Rainfall Measuring Mission (TRMM) Microwave Imager (TMI), launched in November 1997, possesses a 10.65 GHz channel and the Advanced Microwave Scanning Radiometer (AMSR) on both the ADEOS-11 and Earth Observing System AM-1 platforms to be launched in 1999 possesses a 6.9 GHz channel. Aside from issues about interference from vegetation, the coarse resolution of these data will provide considerable challenges pertaining to their application. The resolution of TMI is about 45 km and that of AMSR is about 70 km. These resolutions are grossly inconsistent with the scale of soil moisture processes and the spatial variability of factors that control soil moisture. Scale disparities such as these are forcing us to rethink how we assimilate data of various scales in hydrologic models. Of particular interest is how to assimilate soil moisture data by reconciling the scale disparity between what we can expect from present and future remote sensing measurements of soil moisture and modeling soil moisture processes. It is because of this disparity between the resolution of space-based sensors and the scale of data needed for capturing the spatial variability of soil moisture and related properties that remote sensing of soil moisture has not met with more widespread success. Within a single footprint of current sensors at the wavelengths optimal for this application, in most cases there is enormous heterogeneity in soil moisture created by differences in landcover, soils and topography, as well as variability in antecedent precipitation. It is difficult to interpret the meaning of 'mean' soil moisture under such conditions and even more difficult to apply such a value. Because of the non-linear relationships between near-surface soil moisture and other variables of interest, such as surface energy fluxes and runoff, mean soil moisture has little applicability at such large scales. It is for these reasons that the use of remote sensing in conjunction with a hydrologic model appears to be of benefit in capturing the complete spatial and temporal structure of soil moisture. This paper is Part I of a four-part series describing a method for intermittently assimilating remotely-sensed soil moisture information to improve performance of a distributed land surface hydrology model. The method, summarized in section II, involves the following components, each of which is detailed in the indicated section of the paper or subsequent papers in this series: Forward radiative transfer model methods (section II and Part IV); Use of a Kalman filter to assimilate remotely-sensed soil moisture estimates with the model profile (section II and Part IV); Application of a soil hydrology model to capture the continuous evolution of the soil moisture profile within and below the root zone (section III); Statistical aggregation techniques (section IV and Part II); Disaggregation techniques using a neural network approach (section IV and Part III); and Maximum likelihood and Bayesian algorithms for inversely solving for the soil moisture profile in the upper few cm (Part IV).

Crosson, William L.↗

Evaluating the Toxicity Reduction With Computed Tomographic Ventilation Functional Avoidance Radiation Therapy

Purpose: Computed tomographic (CT) ventilation imaging is a new modality that uses 4-dimensional (4D) CT information to calculate lung ventilation. Although retrospective studies have reported on the reduction in dose to functional lung, no work to our knowledge has been published in which the dosimetric improvements have been translated to a reduction in the probability of pulmonary toxicity. Our work estimates the reduction in toxicity for CT ventilation–based functional avoidance planning. Methods and Materials: Seventy previously treated lung cancer patients who underwent 4DCT imaging were used for the study. CT ventilation maps were calculated with 4DCT deformable image registration and a density change–based algorithm. Pneumonitis was graded on the basis of imaging and clinical presentation. Maximum likelihood methods were used to generate normal tissue complication probability (NTCP) models predicting grade 2 or higher (2+) and grade 3+ pneumonitis as a function of dose (V5 Gy, V10 Gy, V20 Gy, V30 Gy, and mean dose) to functional lung. For 30 patients a functional plan was generated with the goal of reducing dose to the functional lung while meeting Radiation Therapy Oncology Group 0617 constraints. The NTCP models were applied to the functional plans and the clinically used plans to calculate toxicity reduction. Results: By the use of functional avoidance planning, absolute reductions in grade 2+ NTCP of 6.3%, 7.8%, and 4.8% were achieved based on the mean fV20 Gy, fV30 Gy, and mean dose to functional lung metrics, respectively. Absolute grade 3+ NTCP reductions of 3.6%, 4.8%, and 2.4% were achieved with fV20 Gy, fV30 Gy, and mean dose to functional lung. Maximum absolute reductions of 52.3% and 16.4% were seen for grade 2+ and grade 3+ pneumonitis for individual patients. Conclusion: Our study quantifies the possible toxicity reduction from CT ventilation–based functional avoidance planning. Reductions in grades 2+ and 3+ pneumonitis were 7.1% and 4.7% based on mean dose-function metrics, with reductions as high as 52.3% for individual patients. Our work provides seminal data for determining the potential toxicity benefit from incorporating CT ventilation into thoracic treatment planning.

62 RADIOLOGY AND NUCLEAR MEDICINE↗

Maxium-Likelihood Template Matching

In image matching applications such as tracking and stereo matching, it is common to use the sum-of-squared-differences (SSD) measure to determine the best match for an image template.

maximum-likeihood estimation template matching obj↗

RUPRECHT 147: THE OLDEST NEARBY OPEN CLUSTER AS A NEW BENCHMARK FOR STELLAR ASTROPHYSICS

Ruprecht 147 is a hitherto unappreciated open cluster that holds great promise as a standard in fundamental stellar astrophysics. We have conducted a radial velocity survey of astrometric candidates with Lick, Palomar, and MMT observatories and have identified over 100 members, including 5 blue stragglers, 11 red giants, and 5 double-lined spectroscopic binaries (SB2s). We estimate the cluster metallicity from spectroscopic analysis, using Spectroscopy Made Easy (SME), and find it to be [M/H] = +0.07 {+-} 0.03. We have obtained deep CFHT/MegaCam g'r'i'z' photometry and fit Padova isochrones to the (g' - i') and Two Micron All Sky Survey (J - K{sub S} ) color-magnitude diagrams, using the {tau}{sup 2} maximum-likelihood procedure of Naylor, and an alternative method using two-dimensional cross-correlations developed in this work. We find best fits for Padova isochrones at age t = 2.5 {+-} 0.25 Gyr, m - M = 7.35 {+-} 0.1, and A{sub V} = 0.25 {+-} 0.05, with additional uncertainty from the unresolved binary population and possibility of differential extinction across this large cluster. The inferred age is heavily dependent on our choice of stellar evolution model: fitting Dartmouth and PARSEC models yield age parameters of 3 Gyr and 3.25 Gyr, respectively. At {approx}300 pc and {approx}3 Gyr, Ruprecht 147 is by far the oldest nearby star cluster.

79 ASTRONOMY AND ASTROPHYSICS↗

Forest inventory using multistage sampling with probability proportional to size

A multistage sampling technique, with probability proportional to size, for forest volume inventory using remote sensing data is developed and evaluated. The study area is located in the Southeastern Brazil. The LANDSAT 4 digital data of the study area are used in the first stage for automatic classification of reforested areas. Four classes of pine and eucalypt with different tree volumes are classified utilizing a maximum likelihood classification algorithm. Color infrared aerial photographs are utilized in the second stage of sampling. In the third state (ground level) the time volume of each class is determined. The total time volume of each class is expanded through a statistical procedure taking into account all the three stages of sampling. This procedure results in an accurate time volume estimate with a smaller number of aerial photographs and reduced time in field work.

Parada, N. D. J.↗

Poisson vs. Gaussian statistics for sparse X-ray data: Application to the soft X-ray spectrometer

Abstract Reliable results when fitting X-ray data require proper consideration of the statistics involved. We probe the impact of Gaussian versus Poisson statistics at low count levels using both the standard χ2 method and maximum likelihood based on Poisson (C) statistics. The difference is studied and quantified through simulated spectra with known properties. We then test the results through analysis of Mn Kα calibration data taken with the flight spare microcalorimeter for the Hitomi soft X-ray spectrometer. Through comparison with simulations, our results show that the χ2 method tends to give overly optimistic estimates of the detector energy resolution, in particular when there are few counts. Given an energy resolution of ∼5 eV and a line with about 100 photons, the line width becomes ∼10% lower in the χ2 method than in Poisson statistics. This is a consequence of the uncertainties being dominated by counting statistics, and therefore highlights the need to choose the appropriate fit statistic.

79 ASTRONOMY AND ASTROPHYSICS↗

Normal Tissue Complication Probability Estimation by the Lyman-Kutcher-Burman Method Does Not Accurately Predict Spinal Cord Tolerance to Stereotactic Radiosurgery

Purpose: To determine whether normal tissue complication probability (NTCP) analyses of the human spinal cord by use of the Lyman-Kutcher-Burman (LKB) model, supplemented by linear-quadratic modeling to account for the effect of fractionation, predict the risk of myelopathy from stereotactic radiosurgery (SRS). Methods and Materials: From November 2001 to July 2008, 24 spinal hemangioblastomas in 17 patients were treated with SRS. Of the tumors, 17 received 1 fraction with a median dose of 20 Gy (range, 18-30 Gy) and 7 received 20 to 25 Gy in 2 or 3 sessions, with cord maximum doses of 22.7 Gy (range, 17.8-30.9 Gy) and 22.0 Gy (range, 20.2-26.6 Gy), respectively. By use of conventional values for {alpha}/{beta}, volume parameter n, 50% complication probability dose TD{sub 50}, and inverse slope parameter m, a computationally simplified implementation of the LKB model was used to calculate the biologically equivalent uniform dose and NTCP for each treatment. Exploratory calculations were performed with alternate values of {alpha}/{beta} and n. Results: In this study 1 case (4%) of myelopathy occurred. The LKB model using radiobiological parameters from Emami and the logistic model with parameters from Schultheiss overestimated complication rates, predicting 13 complications (54%) and 18 complications (75%), respectively. An increase in the volume parameter (n), to assume greater parallel organization, improved the predictive value of the models. Maximum-likelihood LKB fitting of {alpha}/{beta} and n yielded better predictions (0.7 complications), with n = 0.023 and {alpha}/{beta} = 17.8 Gy. Conclusions: The spinal cord tolerance to the dosimetry of SRS is higher than predicted by the LKB model using any set of accepted parameters. Only a high {alpha}/{beta} value in the LKB model and only a large volume effect in the logistic model with Schultheiss data could explain the low number of complications observed. This finding emphasizes that radiobiological models traditionally used to estimate spinal cord NTCP may not apply to the dosimetry of SRS. Further research with additional NTCP models is needed.

62 RADIOLOGY AND NUCLEAR MEDICINE↗

Poisson vs. Gaussian statistics for sparse X-ray data: Application to the soft X-ray spectrometer

Reliable results when fitting X-ray data require proper consideration of the statistics involved. We probe the impact of Gaussian versus Poisson statistics at low count levels using both the standard χ^(2) method and maximum likelihood based on Poisson (C) statistics. The difference is studied and quantified through simulated spectra with known properties. We then test the results through analysis of Mn Kα calibration data taken with the flight spare microcalorimeter for the Hitomi soft X-ray spectrometer. Through comparison with simulations, our results show that the χ^(2) method tends to give overly optimistic estimates of the detector energy resolution, in particular when there are few counts. Given an energy resolution of ∼5 eV and a line with about 100 photons, the line width becomes ∼10% lower in the χ^(2) method than in Poisson statistics. This is a consequence of the uncertainties being dominated by counting statistics, and therefore highlights the need to choose the appropriate fit statistic.

Shinya Yamada↗

Minimally Informative Prior Distributions for PSA

A salient feature of Bayesian inference is its ability to incorporate information from a variety of sources into the inference model, via the prior distribution (hereafter simply “the prior”). However, over-reliance on old information can lead to priors that dominate new data. Some analysts seek to avoid this by trying to work with a minimally informative prior distribution. Another reason for choosing a minimally informative prior is to avoid the often-voiced criticism of subjectivity in the choice of prior. Minimally informative priors fall into two broad classes: 1) so-called noninformative priors, which attempt to be completely objective, in that the posterior distribution is determined as completely as possible by the observed data, the most well known example in this class being the Jeffreys prior, and 2) priors that are diffuse over the region where the likelihood function is nonnegligible, but that incorporate some information about the parameters being estimated, such as a mean value. In this paper, we compare four approaches in the second class, with respect to their practical implications for Bayesian inference in Probabilistic Safety Assessment (PSA). The most commonly used such prior, the so-called constrained noninformative prior, is a special case of the maximum entropy prior. This is formulated as a conjugate distribution for the most commonly encountered aleatory models in PSA, and is correspondingly mathematically convenient; however, it has a relatively light tail and this can cause the posterior mean to be overly influenced by the prior in updates with sparse data. A more informative prior that is capable, in principle, of dealing more effectively with sparse data is a mixture of conjugate priors. A particular diffuse nonconjugate prior, the logistic-normal, is shown to behave similarly for some purposes. Finally, we review the so-called robust prior. Rather than relying on the mathematical abstraction of entropy, as does the constrained noninformative prior, the robust prior places a heavy-tailed Cauchy prior on the canonical parameter of the aleatory model.

99 GENERAL AND MISCELLANEOUS↗

Parameter identification of linear discrete stochastic systems with time delays

An identification algorithm that uses the maximum likelihood technique to identify the unknown time delays, plant parameters, and noise covariances of linear discrete stochastic systems is presented. Cases of additive white noise and colored measurement noises are considered. The likelihood function is evaluated using either a minimum-variance (Kalman) filter or a minimal-order observer. The Kalman filter is used in the identification algorithm to provide minimum-variance estimates. The minimal-order observer is a lower-dimensional and computationally simpler filter, and is advantageous especially for systems with long delays. It provides a less optimal solution to the minimum-mean-square state estimation problem. The colored-noise observer algorithm has the disadvantage of having to compute an extra error covariance matrix of lower order.

Wong, E. C.↗

Determination of mean rainfall from the Special Sensor Microwave/Imager (SSM/I) using a mixed lognormal distribution

Global estimates of monthly, seasonal, and annual oceanic rainfall are computed for a period of one year using data from the Special Sensor Microwave/Imager (SSM/I). Instantaneous rainfall estimates are derived from brightness temperature values obtained from the satellite data using the Hughes D-matrix algorithm. The instantaneous rainfall estimates are stored in 1 deg square bins over the global oceans for each month. A mixed probability distribution combining a lognormal distribution describing the positive rainfall values and a spike at zero describing the observations indicating no rainfall is used to compute mean values. The resulting data for the period of interest are fitted to a lognormal distribution by using a maximum-likelihood. Mean values are computed for the mixed distribution and qualitative comparisons with published historical results as well as quantitative comparisons with corresponding in situ raingage data are performed.

Berg, Wesley↗

Comparison of Fault Detection Algorithms for Real-time Diagnosis in Large-Scale System

In this paper, we present a review of different real-time capable algorithms to detect and isolate component failures in large-scale systems in the presence of inaccurate test results. A sequence of imperfect test results (as a row vector of I's and O's) are available to the algorithms. In this case, the problem is to recover the uncorrupted test result vector and match it to one of the rows in the test dictionary, which in turn will isolate the faults. In order to recover the uncorrupted test result vector, one needs the accuracy of each test. That is, its detection and false alarm probabilities are required. In this problem, their true values are not known and, therefore, have to be estimated online. Other major aspects in this problem are the large-scale nature and the real-time capability requirement. Test dictionaries of sizes up to 1000 x 1000 are to be handled. That is, results from 1000 tests measuring the state of 1000 components are available. However, at any time, only 10-20% of the test results are available. Then, the objective becomes the real-time fault diagnosis using incomplete and inaccurate test results with online estimation of test accuracies. It should also be noted that the test accuracies can vary with time --- one needs a mechanism to update them after processing each test result vector. Using Qualtech's TEAMS-RT (system simulation and real-time diagnosis tool), we test the performances of 1) TEAMSAT's built-in diagnosis algorithm, 2) Hamming distance based diagnosis, 3) Maximum Likelihood based diagnosis, and 4) HidderMarkov Model based diagnosis.

Kirubarajan, Thiagalingam↗

The 2.4 μm Galaxy Luminosity Function As Measured Using WISE. I. Measurement Techniques

The astronomy community has at its disposal a large back catalog of public spectroscopic galaxy redshift surveys that can be used for the measurement of luminosity functions (LFs). Utilizing the back catalog with new photometric surveys to maximum efficiency requires modeling the color selection bias imposed on the selection of target galaxies by flux limits at multiple wavelengths. The likelihood derived herein can address, in principle, all possible color selection biases through the use of a generalization of the LF, Φ(L), over the space of all spectra: the spectro-luminosity functional, Ψ[L{sub ν}]. It is, therefore, the first estimator capable of simultaneously analyzing multiple redshift surveys in a consistent way. We also propose a new way of parametrizing the evolution of the classic Schechter function parameters, L {sub ⋆} and ϕ {sub ⋆}, that improves both the physical realism and statistical performance of the model. The techniques derived in this paper are used in a companion paper by Lake et al. to measure the LF of galaxies at the rest-frame wavelength of 2.4 μm using the Widefield Infrared Survey Explorer (WISE).

79 ASTRONOMY AND ASTROPHYSICS↗

Extreme-Scale Bayesian Inference for Uncertainty Quantification of Complex Simulations

Uncertainty quantification (UQ)—that is, quantifying uncertainties in complex mathematical models and their large-scale computational implementations—is widely viewed as one of the outstanding challenges facing the field of CS&E over the coming decade. The EUREKA project set to address the most difficult class of UQ problems: those for which both the underlying PDE model as well as the uncertain parameters are of extreme scale. In the project we worked on these extreme-scale challenges in the following four areas: 1. Scalable parallel algorithms for sampling and characterizing the posterior distribution that exploit the structure of the underlying PDEs and parameter-to-observable map. These include structure-exploiting versions of the randomized maximum likelihood method, which aims to overcome the intractability of employing conventional MCMC methods for solving extreme-scale Bayesian inversion problems by appealing to and adapting ideas from large-scale PDE-constrained optimization, which have been very successful at exploring high-dimensional spaces. 2. Scalable parallel algorithms for construction of prior and likelihood functions based on learning methods and non-parametric density estimation. Constructing problem-specific priors remains a critical challenge in Bayesian inference, and more so in high dimensions. Another challenge is construction of likelihood functions that capture unmodeled couplings between observations and parameters. We will create parallel algorithms for non-parametric density estimation using high dimensional N-body methods and combine them with supervised learning techniques for the construction of priors and likelihood functions. 3. Bayesian inadequacy models, which augment physics models with stochastic models that represent their imperfections. The success of the Bayesian inference framework depends on the ability to represent the uncertainty due to imperfections of the mathematical model of the phenomena of interest. This is a central challenge in UQ, especially for large-scale models. We propose to develop the mathematical tools to address these challenges in the context of extreme-scale problems. 4. Parallel scalable algorithms for Bayesian optimal experimental design (OED). Bayesian inversion yields quantified uncertainties in the model parameters, which can be propagated forward through the model to yield uncertainty in outputs of interest. This opens the way for designing new experiments to reduce the uncertainties in the model parameters and model predictions. Such experimental design problems have been intractable for large-scale problems using conventional methods; we will create OED algorithms that exploit the structure of the PDE model and the parameter-to-output map to overcome these challenges. Parallel algorithms for these four problems were created, analyzed, prototyped, implemented, tuned, and scaled up for leading-edge supercomputers, including UT-Austin’s own 10 petaflops Stampede system, ANL’s Mira system, and ORNL’s Titan system. While our focus is on fundamental mathematical/computational methods and algorithms, we will assess our methods on model problems derived from several DOE mission applications, including multiscale mechanics and ice sheet dynamics.

97 MATHEMATICS AND COMPUTING↗

The design of a turboshaft speed governor using modern control techniques

The objectives of this program were: to verify the model of off schedule compressor variable geometry in the T700 turboshaft engine nonlinear model; to evaluate the use of the pseudo-random binary noise (PRBN) technique for obtaining engine frequency response data; and to design a high performance power turbine speed governor using modern control methods. Reduction of T700 engine test data generated at NASA-Lewis indicated that the off schedule variable geometry effects were accurate as modeled. Analysis also showed that the PRBN technique combined with the maximum likelihood model identification method produced a Bode frequency response that was as accurate as the response obtained from standard sinewave testing methods. The frequency response verified the accuracy of linear models consisting of engine partial derivatives and used for design. A power turbine governor was designed using the Linear Quadratic Regulator (LQR) method of full state feedback control. A Kalman filter observer was used to estimate helicopter main rotor blade velocity. Compared to the baseline T700 power turbine speed governor, the LQR governor reduced droop up to 25 percent for a 490 shaft horsepower transient in 0.1 sec simulating a wind gust, and up to 85 percent for a 700 shaft horsepower transient in 0.5 sec simulating a large collective pitch angle transient.

Delosreyes, G.↗

Flight-Determined, Subsonic, Lateral-Directional Stability and Control Derivatives of the Thrust-Vectoring F-18 High Angle of Attack Research Vehicle (HARV), and Comparisons to the Basic F-18 and Predicted Derivatives

The subsonic, lateral-directional, stability and control derivatives of the thrust-vectoring F-1 8 High Angle of Attack Research Vehicle (HARV) are extracted from flight data using a maximum likelihood parameter identification technique. State noise is accounted for in the identification formulation and is used to model the uncommanded forcing functions caused by unsteady aerodynamics. Preprogrammed maneuvers provided independent control surface inputs, eliminating problems of identifiability related to correlations between the aircraft controls and states. The HARV derivatives are plotted as functions of angles of attack between 10deg and 70deg and compared to flight estimates from the basic F-18 aircraft and to predictions from ground and wind tunnel tests. Unlike maneuvers of the basic F-18 aircraft, the HARV maneuvers were very precise and repeatable, resulting in tightly clustered estimates with small uncertainty levels. Significant differences were found between flight and prediction; however, some of these differences may be attributed to differences in the range of sideslip or input amplitude over which a given derivative was evaluated, and to differences between the HARV external configuration and that of the basic F-18 aircraft, upon which most of the prediction was based. Some HARV derivative fairings have been adjusted using basic F-18 derivatives (with low uncertainties) to help account for differences in variable ranges and the lack of HARV maneuvers at certain angles of attack.

Iliff, Kenneth W.↗

Adaptive hyperparameter updating for training restricted Boltzmann machines on quantum annealers

Restricted Boltzmann Machines (RBMs) have been proposed for developing neural networks for a variety of unsupervised machine learning applications such as image recognition, drug discovery, and materials design. The Boltzmann probability distribution is used as a model to identify network parameters by optimizing the likelihood of predicting an output given hidden states trained on available data. Training such networks often requires sampling over a large probability space that must be approximated during gradient based optimization. Quantum annealing has been proposed as a means to search this space more efficiently which has been experimentally investigated on D-Wave hardware. D-Wave implementation requires selection of an effective inverse temperature or hyperparameter (β) within the Boltzmann distribution which can strongly influence optimization. Here, we show how this parameter can be estimated as a hyperparameter applied to D-Wave hardware during neural network training by maximizing the likelihood or minimizing the Shannon entropy. We find both methods improve training RBMs based upon D-Wave hardware experimental validation on an image recognition problem. Neural network image reconstruction errors are evaluated using Bayesian uncertainty analysis which illustrate more than an order magnitude lower image reconstruction error using the maximum likelihood over manually optimizing the hyperparameter. The maximum likelihood method is also shown to out-perform minimizing the Shannon entropy for image reconstruction.

97 MATHEMATICS AND COMPUTING↗