Search NASA⌕ Search

SEARCH · Search NASA

Results for “Statistical error”

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 1,063 records · Page 59

Global and Local Distortion Inference During Embedded Zerotree Wavelet Decompression

This paper presents algorithms for inferring global and spatially local estimates of the squared-error distortion measures for the Embedded Zerotree Wavelet (EZW) image compression algorithm. All distortion estimates are obtained at the decoder without significantly compromising EZW's rate-distortion performance. Two methods are given for propagating distortion estimates from the wavelet domain to the spatial domain, thus giving individual estimates of distortion for each pixel of the decompressed image. These local distortion estimates seem to provide only slight improvement in the statistical characterization of EZW compression error relative to the global measure, unless actual squared errors are propagated. However, they provide qualitative information about the asymptotic nature of the error that may be helpful in wavelet filter selection for low bit rate applications.

Huber, A. Kris↗

SU-F-T-362: Quantification and Modelling of the Ionization Chamber Simulation Effective Points On Monaco Treatment Planning System

Purpose: Because of statistical noise in Monte Carlo dose calculations, effective point doses may not be accurate. Volume spheres are useful for evaluating dose in Monte Carlo plans, which have an inherent statistical uncertainty.We use a user-defined sphere volume instead of a point, take sphere sampling around effective point make the dose statistics to decrease the stochastic errors. Methods: Direct dose measurements were made using a 0.125cc Semiflex ion chamber (IC) 31010 isocentrically placed in the center of a homogeneous Cylindric sliced RW3 phantom (PTW, Germany).In the scanned CT phantom series the sensitive volume length of the IC (6.5mm) were delineated and defined the isocenter as the simulation effective points. All beams were simulated in Monaco in accordance to the measured model. In our simulation using 2mm voxels calculation grid spacing and choose calculate dose to medium and request the relative standard deviation ≤0.5%. Taking three different assigned IC over densities (air electron density(ED) as 0.01g/cm3 default CT scanned ED and Esophageal lumen ED 0.21g/cm3) were tested at different sampling sphere radius (2.5, 2, 1.5 and 1 mm) statistics dose were compared with the measured does. Results: The results show that in the Monaco TPS for the IC using Esophageal lumen ED 0.21g/cm3 and sampling sphere radius 1.5mm the statistical value is the best accordance with the measured value, the absolute average percentage deviation is 0.49%. And when the IC using air electron density(ED) as 0.01g/cm3 and default CT scanned EDthe recommented statistical sampling sphere radius is 2.5mm, the percentage deviation are 0.61% and 0.70%, respectivly. Conclusion: In Monaco treatment planning system for the ionization chamber 31010 recommend air cavity using ED 0.21g/cm3 and sampling 1.5mm sphere volume instead of a point dose to decrease the stochastic errors. Funding Support No.C201505006.

60 APPLIED LIFE SCIENCES↗

TRUE MASSES OF RADIAL-VELOCITY EXOPLANETS

We study the task of estimating the true masses of known radial-velocity (RV) exoplanets by means of direct astrometry on coronagraphic images to measure the apparent separation between exoplanet and host star. Initially, we assume perfect knowledge of the RV orbital parameters and that all errors are due to photon statistics. We construct design reference missions for four missions currently under study at NASA: EXO-S and WFIRST-S, with external star shades for starlight suppression, EXO-C and WFIRST-C, with internal coronagraphs. These DRMs reveal extreme scheduling constraints due to the combination of solar and anti-solar pointing restrictions, photometric and obscurational completeness, image blurring due to orbital motion, and the “nodal effect,” which is the independence of apparent separation and inclination when the planet crosses the plane of the sky through the host star. Next, we address the issue of nonzero uncertainties in RV orbital parameters by investigating their impact on the observations of 21 single-planet systems. Except for two—GJ 676 A b and 16 Cyg B b, which are observable only by the star-shade missions—we find that current uncertainties in orbital parameters generally prevent accurate, unbiased estimation of true planetary mass. For the coronagraphs, WFIRST-C and EXO-C, the most likely number of good estimators of true mass is currently zero. For the star shades, EXO-S and WFIRST-S, the most likely numbers of good estimators are three and four, respectively, including GJ 676 A b and 16 Cyg B b. We expect that uncertain orbital elements currently undermine all potential programs of direct imaging and spectroscopy of RV exoplanets.

79 ASTRONOMY AND ASTROPHYSICS↗

Cluster Mass Calibration at High Redshift: HST Weak Lensing Analysis of 13 Distant Galaxy Clusters from the South Pole Telescope Sunyaev-Zel’dovich Survey

We present an HST/Advanced Camera for Surveys (ACS) weak gravitational lensing analysis of 13 massive high-redshift (z(median) = 0.88) galaxy clusters discovered in the South Pole Telescope (SPT) Sunyaev-Zel'dovich Survey. This study is part of a larger campaign that aims to robustly calibrate mass-observable scaling relations over a wide range in redshift to enable improved cosmological constraints from the SPT cluster sample. We introduce new strategies to ensure that systematics in the lensing analysis do not degrade constraints on cluster scaling relations significantly. First, we efficiently remove cluster members from the source sample by selecting very blue galaxies in V - I colour. Our estimate of the source redshift distribution is based on Cosmic Assembly Near-infrared Deep Extragalactic Legacy Survey (CANDELS) data, where we carefully mimic the source selection criteria of the cluster fields. We apply a statistical correction for systematic photometric redshift errors as derived from Hubble Ultra Deep Field data and verified through spatial cross-correlations. We account for the impact of lensing magnification on the source redshift distribution, finding that this is particularly relevant for shallower surveys. Finally, we account for biases in the mass modelling caused by miscentring and uncertainties in the concentration-mass relation using simulations. In combination with temperature estimates from Chandra we constrain the normalization of the mass-temperature scaling relation ln (E(z) M-500c/10(14)M(circle dot)) = A + 1.5ln (kT/7.2 keV) to A = 1.81(-0.14)(+0.24)(stat.)+/- 0.09(sys.), consistent with self-similar redshift evolution when compared to lower redshift samples. Additionally, the lensing data constrain the average concentration of the clusters to c(200c) = 5.6(-1.8)(+3.7).

cosmology: observations↗

Seeing through noise in power laws

Despite widespread claims of power laws across the natural and social sciences, evidence in data is often equivocal. Modern data and statistical methods reject even classic power laws such as Pareto’s law of wealth and the Gutenberg–Richter law for earthquake magnitudes. We show that the maximum-likelihood estimators and Kolmogorov–Smirnov (K-S) statistics in widespread use are unexpectedly sensitive to ubiquitous errors in data such as measurement noise, quantization noise, heaping and censorship of small values. This sensitivity causes spurious rejection of power laws and biases parameter estimates even in arbitrarily large samples, which explains inconsistencies between theory and data. We show that logarithmic binning by powers of λ > 1 attenuates these errors in a manner analogous to noise averaging in normal statistics and that λ thereby tunes a trade-off between accuracy and precision in estimation. Binning also removes potentially misleading within-scale information while preserving information about the shape of a distribution over powers of λ, and we show that some amount of binning can improve sensitivity and specificity of K-S tests without any cost, while more extreme binning tunes a trade-off between sensitivity and specificity. We therefore advocate logarithmic binning as a simple essential step in power-law inference.

29 ENERGY PLANNING, POLICY, AND ECONOMY↗

Seeing through noise in power laws

Despite widespread claims of power laws across the natural and social sciences, evidence in data is often equivocal. Modern data and statistical methods reject even classic power laws such as Pareto’s law of wealth and the Gutenberg–Richter law for earthquake magnitudes. We show that the maximum-likelihood estimators and Kolmogorov–Smirnov (K-S) statistics in widespread use are unexpectedly sensitive to ubiquitous errors in data such as measurement noise, quantization noise, heaping and censorship of small values. This sensitivity causes spurious rejection of power laws and biases parameter estimates even in arbitrarily large samples, which explains inconsistencies between theory and data. We show that logarithmic binning by powers of λ > 1 attenuates these errors in a manner analogous to noise averaging in normal statistics and that λ thereby tunes a trade-off between accuracy and precision in estimation. Binning also removes potentially misleading within-scale information while preserving information about the shape of a distribution over powers of λ, and we show that some amount of binning can improve sensitivity and specificity of K-S tests without any cost, while more extreme binning tunes a trade-off between sensitivity and specificity. We therefore advocate logarithmic binning as a simple essential step in power-law inference.

97 MATHEMATICS AND COMPUTING↗

Adjoint Error Estimation for Linear Advection

An a posteriori error formula is described when a statistical measurement of the solution to a hyperbolic conservation law in 1D is estimated by finite volume approximations. This is accomplished using adjoint error estimation. In contrast to previously studied methods, the adjoint problem is divorced from the finite volume method used to approximate the forward solution variables. An exact error formula and computable error estimate are derived based on an abstractly defined approximation of the adjoint solution. This framework allows the error to be computed to an arbitrary accuracy given a sufficiently well resolved approximation of the adjoint solution. The accuracy of the computable error estimate provably satisfies an a priori error bound for sufficiently smooth solutions of the forward and adjoint problems. The theory does not currently account for discontinuities. Computational examples are provided that show support of the theory for smooth solutions. The application to problems with discontinuities is also investigated computationally.

97 MATHEMATICS AND COMPUTING↗

The beam filling error in the Nimbus 5 electronically scanning microwave radiometer observations of Global Atlantic Tropical Experiment rainfall

A comparison of rain rates retrieved from the Nimbus 5 electronically scanning microwave radiometer brightness temperatures and observed from shipboard radars during the Global Atlantic Tropical Experiment (GATE) phase I shows that the beam filling error is the major source of discrepancy between the two. When averaged over a large scene (the GATE radar array, 400 km in diameter), the beam filling error is quite stable, being 50 percent of the observed rain rate. This suggests the simple procedure of multiplying retrieved rain rates by 2 (correction factor). A statistical model of the beam filling error is developed by envisioning an idealized instrument field-of-view that encompasses an entire gamma distribution of rain rates. A modeled correction factor near 2 is found for rain rate and temperature characteristics consistent with GATE conditions. The statistical model also suggests that the correction factor varies from 1.5 to 2.5 for suppressed to enhanced tropical convective regimes, and decreases to 1.5 as the freezing level and average depth of the rain column decreases to 2.5 km.

Short, David A.↗

Automatic recognition of intermittent failures - An experimental study of field data

A methodology is proposed for recognizing the symptoms of persistent problems in large systems. The system error rate is used to identify the error states among which relationships may exist. Statistical techniques are used to validate and quantify the strength of the relationship among these error states. As input, the approach takes the raw error logs containing a single entry for each error that is detected as an isolated event. As output, it produces a list of symptoms that characterize persistent errors. Thus, given a failure, it is determined whether the failure is an intermittent manifestation of a common fault or whether it is an isolated (transient) incident. The technique is shown to work on two CYBER systems and an IBM 3081 multiprocessor system. Comparisons to real failure/repair information obtained from field engineers show that, in about 85 percent of the cases, the error symptoms recognized by this approach correspond to real problems. The remaining 15 percent of the cases, although not directly supported by field data, are confirmed as being valid problems.

Iyer, Ravishankar K.↗

Liver segmentation in contrast enhanced CT data using graph cuts and interactive 3D segmentation refinement methods

Purpose: Liver segmentation is an important prerequisite for the assessment of liver cancer treatment options like tumor resection, image-guided radiation therapy (IGRT), radiofrequency ablation, etc. The purpose of this work was to evaluate a new approach for liver segmentation. Methods: A graph cuts segmentation method was combined with a three-dimensional virtual reality based segmentation refinement approach. The developed interactive segmentation system allowed the user to manipulate volume chunks and/or surfaces instead of 2D contours in cross-sectional images (i.e, slice-by-slice). The method was evaluated on twenty routinely acquired portal-phase contrast enhanced multislice computed tomography (CT) data sets. An independent reference was generated by utilizing a currently clinically utilized slice-by-slice segmentation method. After 1 h of introduction to the developed segmentation system, three experts were asked to segment all twenty data sets with the proposed method. Results: Compared to the independent standard, the relative volumetric segmentation overlap error averaged over all three experts and all twenty data sets was 3.74%. Liver segmentation required on average 16 min of user interaction per case. The calculated relative volumetric overlap errors were not found to be significantly different [analysis of variance (ANOVA) test, p = 0.82] between experts who utilized the proposed 3D system. In contrast, the time required by each expert for segmentation was found to be significantly different (ANOVA test, p = 0.0009). Major differences between generated segmentations and independent references were observed in areas were vessels enter or leave the liver and no accepted criteria for defining liver boundaries exist. In comparison, slice-by-slice based generation of the independent standard utilizing a live wire tool took 70.1 min on average. A standard 2D segmentation refinement approach applied to all twenty data sets required on average 38.2 min of user interaction and resulted in statistically not significantly different segmentation error indices (ANOVA test, significance level of 0.05). Conclusions: All three experts were able to produce liver segmentations with low error rates. User interaction time savings of up to 71% compared to a 2D refinement approach demonstrate the utility and potential of our approach. The system offers a range of different tools to manipulate segmentation results, and some users might benefit from a longer learning phase to develop efficient segmentation refinement strategies. The presented approach represents a generally applicable segmentation approach that can be applied to many medical image segmentation problems.

62 RADIOLOGY AND NUCLEAR MEDICINE↗

A semiempirical linear model of indirect, flat-panel x-ray detectors

Purpose: It is important to understand signal and noise transfer in the indirect, flat-panel x-ray detector when developing and optimizing imaging systems. For optimization where simulating images is necessary, this study introduces a semiempirical model to simulate projection images with user-defined x-ray fluence interaction. Methods: The signal and noise transfer in the indirect, flat-panel x-ray detectors is characterized by statistics consistent with energy-integration of x-ray photons. For an incident x-ray spectrum, x-ray photons are attenuated and absorbed in the x-ray scintillator to produce light photons, which are coupled to photodiodes for signal readout. The signal mean and variance are linearly related to the energy-integrated x-ray spectrum by empirically determined factors. With the known first- and second-order statistics, images can be simulated by incorporating multipixel signal statistics and the modulation transfer function of the imaging system. To estimate the semiempirical input to this model, 500 projection images (using an indirect, flat-panel x-ray detector in the breast CT system) were acquired with 50-100 kilovolt (kV) x-ray spectra filtered with 0.1-mm tin (Sn), 0.2-mm copper (Cu), 1.5-mm aluminum (Al), or 0.05-mm silver (Ag). The signal mean and variance of each detector element and the noise power spectra (NPS) were calculated and incorporated into this model for accuracy. Additionally, the modulation transfer function of the detector system was physically measured and incorporated in the image simulation steps. For validation purposes, simulated and measured projection images of air scans were compared using 40 kV/0.1-mm Sn, 65 kV/0.2-mm Cu, 85 kV/1.5-mm Al, and 95 kV/0.05-mm Ag. Results: The linear relationship between the measured signal statistics and the energy-integrated x-ray spectrum was confirmed and incorporated into the model. The signal mean and variance factors were linearly related to kV for each filter material (r{sup 2} of signal mean to kV: 0.91, 0.93, 0.86, and 0.99 for 0.1-mm Sn, 0.2-mm Cu, 1.5-mm Al, and 0.05-mm Ag, respectively; r{sup 2} of signal variance to kV: 0.99 for all four filters). The comparison of the signal and noise (mean, variance, and NPS) between the simulated and measured air scan images suggested that this model was reasonable in predicting accurate signal statistics of air scan images using absolute percent error. Overall, the model was found to be accurate in estimating signal statistics and spatial correlation between the detector elements of the images acquired with indirect, flat-panel x-ray detectors. Conclusions: The semiempirical linear model of the indirect, flat-panel x-ray detectors was described and validated with images of air scans. The model was found to be a useful tool in understanding the signal and noise transfer within indirect, flat-panel x-ray detector systems.

46 INSTRUMENTATION RELATED TO NUCLEAR SCIENCE AND ↗

Coordinated Ramping Product and Regulation Reserve Procurements in CAISO and MISO using Multi-Scale Probabilistic Solar Power Forecasts (Pro2R)

How can probabilistic solar forecasts lower costs and improve reliability for independent system operator (ISO) markets? We tackle this question in three steps. First, we enhance an existing solar forecasting system to provide well-calibrated hours-ahead probabilistic forecasts. We then relate the degree of uncertainty in those forecasts to error distributions for net load ramps for the California ISO (CAISO) using statistical and machine learning methods. Projected net load errors conditioned on solar uncertainty are translated into flexible ramp requirements that therefore reflect real-time meteorological and solar conditions, improving on typical ISO procedures. Finally, a multi-period look-ahead production cost model quantifies how conditional ramp requirements can a) decrease operating costs by lowering requirements compared to often conservative unconditional methods, and b) reduce generation scarcity events and consequently improve reliability by increasing flexibility requirements at times when unconditional forecast-based requirements understate actual ramp uncertainty. In addition to the products just described (quantification of solar uncertainty, its translation into requirements for ramp capability product, and quantification of the benefits of more accurate ramp requirements), this project also developed a visualization system that alerts system operators of ramp and uncertainty conditions within the network based on solar forecasts. The system is called Resource Forecast and Ramp Visualization for Situational Awareness (RaVIS). These four products represent significant advances in the state-of-the-art of probabilistic solar forecasting, development of weather-informed reserve requirements, production costing methods for estimating the benefits of more accurate reserve requirements, and visualization of system status, respectively. Yet the products are also practical and can be immediately implemented, potentially enabling system operators to save millions of dollars in ramp product procurement costs per year.

14 SOLAR ENERGY↗

WE-DE-201-02: A Statistical Analysis Tool for Plan Quality Verification in HDR Brachytherapy Forward Planning for Cervix Cancer

Purpose: High Dose Rate (HDR) brachytherapy forward planning is principally an iterative process; hence, plan quality is affected by planners’ experiences and limited planning time. Thus, this may lead to sporadic errors and inconsistencies in planning. A statistical tool based on previous approved clinical treatment plans would help to maintain the consistency of planning quality and improve the efficiency of second checking. Methods: An independent dose calculation tool was developed from commercial software. Thirty-three previously approved cervical HDR plans with the same prescription dose (550cGy), applicator type, and treatment protocol were examined, and ICRU defined reference point doses (bladder, vaginal mucosa, rectum, and points A/B) along with dwell times were collected. Dose calculation tool then calculated appropriate range with a 95% confidence interval for each parameter obtained, which would be used as the benchmark for evaluation of those parameters in future HDR treatment plans. Model quality was verified using five randomly selected approved plans from the same dataset. Results: Dose variations appears to be larger at the reference point of bladder and mucosa as compared with rectum. Most reference point doses from verification plans fell between the predicted range, except the doses of two points of rectum and two points of reference position A (owing to rectal anatomical variations & clinical adjustment in prescription points, respectively). Similar results were obtained for tandem and ring dwell times despite relatively larger uncertainties. Conclusion: This statistical tool provides an insight into clinically acceptable range of cervical HDR plans, which could be useful in plan checking and identifying potential planning errors, thus improving the consistency of plan quality.

60 APPLIED LIFE SCIENCES↗

In Flight Calibration of the Magnetospheric Multiscale Mission Fast Plasma Investigation

The Fast Plasma Investigation (FPI) on the Magnetospheric Multiscale mission (MMS) combines data from eight spectrometers, each with four deflection states, into a single map of the sky. Any systematic discontinuity, artifact, noise source, etc. present in this map may be incorrectly interpreted as legitimate data and incorrect conclusions reached. For this reason it is desirable to have all spectrometers return the same output for a given input, and for this output to be low in noise sources or other errors. While many missions use statistical analyses of data to calibrate instruments in flight, this process is difficult with FPI for two reasons: 1. Only a small fraction of high resolution data is downloaded to the ground due to bandwidth limitations and 2: The data that is downloaded is, by definition, scientifically interesting and therefore not ideal for calibration. FPI uses a suite of new tools to calibrate in flight. A new method for detection system ground calibration has been developed involving sweeping the detection threshold to fully define the pulse height distribution. This method has now been extended for use in flight as a means to calibrate MCP voltage and threshold (together forming the operating point) of the Dual Electron Spectrometers (DES) and Dual Ion Spectrometers (DIS). A method of comparing higher energy data (which has low fractional voltage error) to lower energy data (which has a higher fractional voltage error) will be used to calibrate the high voltage outputs. Finally, a comparison of pitch angle distributions will be used to find remaining discrepancies among sensors.

Detection-system↗

In Flight Calibration of the Magnetospheric Multiscale Mission Fast Plasma Investigation

The Fast Plasma Investigation (FPI) on the Magnetospheric Multiscale mission (MMS) combines data from eight spectrometers, each with four deflection states, into a single map of the sky. Any systematic discontinuity, artifact, noise source, etc. present in this map may be incorrectly interpreted as legitimate data and incorrect conclusions reached. For this reason it is desirable to have all spectrometers return the same output for a given input, and for this output to be low in noise sources or other errors. While many missions use statistical analyses of data to calibrate instruments in flight, this process is insufficient with FPI for two reasons: 1. Only a small fraction of high resolution data is downloaded to the ground due to bandwidth limitations and 2: The data that is downloaded is, by definition, scientifically interesting and therefore not ideal for calibration. FPI uses a suite of new tools to calibrate in flight. A new method for detection system ground calibration has been developed involving sweeping the detection threshold to fully define the pulse height distribution. This method has now been extended for use in flight as a means to calibrate MCP voltage and threshold (together forming the operating point) of the Dual Electron Spectrometers (DES) and Dual Ion Spectrometers (DIS). A method of comparing higher energy data (which has low fractional voltage error) to lower energy data (which has a higher fractional voltage error) will be used to calibrate the high voltage outputs. Finally, a comparison of pitch angle distributions will be used to find remaining discrepancies among sensors.

Heliophysics↗

Magnetic Resonance Imaging Assessment of Spinal Cord and Cauda Equina Motion in Supine Patients With Spinal Metastases Planned for Spine Stereotactic Body Radiation Therapy

Purpose: To assess motion of the spinal cord and cauda equina, which are critical neural tissues (CNT), which is important when evaluating the planning organ-at-risk margin required for stereotactic body radiation therapy. Methods and Materials: We analyzed CNT motion in 65 patients with spinal metastases (11 cervical, 39 thoracic, and 24 lumbar spinal segments) in the supine position using dynamic axial and sagittal magnetic resonance imaging (dMRI, 3T Verio, Siemens) over a 137-second interval. Motion was segregated according to physiologic cardiorespiratory oscillatory motion (characterized by the average root mean square deviation) and random bulk shifts associated with gross patient motion (characterized by the range). Displacement was evaluated in the anteroposterior (AP), lateral (LR), and superior-inferior (SI) directions by use of a correlation coefficient template matching algorithm, with quantification of random motion measure error over 3 separate trials. Statistical significance was defined according to P<.05. Results: In the AP, LR, and SI directions, significant oscillatory motion was observed in 39.2%, 35.1%, and 10.8% of spinal segments, respectively, and significant bulk motions in all cases. The median oscillatory CNT motions in the AP, LR, and SI directions were 0.16 mm, 0.17 mm, and 0.44 mm, respectively, and the maximal statistically significant oscillatory motions were 0.39 mm, 0.41 mm, and 0.77 mm, respectively. The median bulk displacements in the AP, LR, and SI directions were 0.51 mm, 0.59 mm, and 0.66 mm, and the maximal statistically significant displacements were 2.21 mm, 2.87 mm, and 3.90 mm, respectively. In the AP, LR, and SI directions, bulk displacements were greater than 1.5 mm in 5.4%, 9.0%, and 14.9% of spinal segments, respectively. No significant differences in axial motion were observed according to cord level or cauda equina. Conclusions: Oscillatory CNT motion was observed to be relatively minor. Our results support the importance of controlling bulk patient motion and the practice of applying a planning organ-at-risk margin.

62 RADIOLOGY AND NUCLEAR MEDICINE↗

Subcell resolution in simplex stochastic collocation for spatial discontinuities

Subcell resolution has been used in the Finite Volume Method (FVM) to obtain accurate approximations of discontinuities in the physical space. Stochastic methods are usually based on local adaptivity for resolving discontinuities in the stochastic dimensions. However, the adaptive refinement in the probability space is ineffective in the non-intrusive uncertainty quantification framework, if the stochastic discontinuity is caused by a discontinuity in the physical space with a random location. The dependence of the discontinuity location in the probability space on the spatial coordinates then results in a staircase approximation of the statistics, which leads to first-order error convergence and an underprediction of the maximum standard deviation. To avoid these problems, we introduce subcell resolution into the Simplex Stochastic Collocation (SSC) method for obtaining a truly discontinuous representation of random spatial discontinuities in the interior of the cells discretizing the probability space. The presented SSC–SR method is based on resolving the discontinuity location in the probability space explicitly as function of the spatial coordinates and extending the stochastic response surface approximations up to the predicted discontinuity location. The applications to a linear advection problem, the inviscid Burgers’ equation, a shock tube problem, and the transonic flow over the RAE 2822 airfoil show that SSC–SR resolves random spatial discontinuities with multiple stochastic and spatial dimensions accurately using a minimal number of samples.

71 CLASSICAL AND QUANTUM MECHANICS, GENERAL PHYSIC↗

Correcting evaluation bias of relational classifiers with network cross validation

Recently, a number of modeling techniques have been developed for data mining and machine learning in relational and network domains where the instances are not independent and identically distributed (i.i.d.). These methods specifically exploit the statistical dependencies among instances in order to improve classification accuracy. However, there has been little focus on how these same dependencies affect our ability to draw accurate conclusions about the performance of the models. More specifically, the complex link structure and attribute dependencies in relational data violate the assumptions of many conventional statistical tests and make it difficult to use these tests to assess the models in an unbiased manner. In this work, we examine the task of within-network classification and the question of whether two algorithms will learn models that will result in significantly different levels of performance. We show that the commonly used form of evaluation (paired t-test on overlapping network samples) can result in an unacceptable level of Type I error. Furthermore, we show that Type I error increases as (1) the correlation among instances increases and (2) the size of the evaluation set increases (i.e., the proportion of labeled nodes in the network decreases). Lastly, we propose a method for network cross-validation that combined with paired t-tests produces more acceptable levels of Type I error while still providing reasonable levels of statistical power (i.e., 1–Type II error).

97 MATHEMATICS AND COMPUTING↗