Reduced Data Set Research Articles

Vision system development by machine learning: Mashing assessment in brewing

This article describes the development of a machine vision application for automatic process assessment by image analysis and machine learning. The system is required to differentiate between the various stages of a mashing process and determine the termination point. A large number of histograms, Haralick and Gabor features (835) were extracted from 275 training images. Three feature selection algorithms - wrapper, consistency filter, and correlation filter - were then applied to the training data, resulting in feature sets of size 29, 15, and 11, respectively. A number of decision tree, rule induction, and nearest neighbor classification algorithms were then applied to the reduced data set. For discriminating seven stages of the mashing process, the highest accuracy obtained was 71.6%. For the binary problem of differentiating the finished state from all of the other states the accuracy was 92.0%. This accuracy is good enough for deployment. The results indicate that using a large library of features and machine-learning methods for removing redundant features can significantly reduce development times for vision systems by eliminating the time-consuming manual search for the best discriminating features.

Impact of variation in length of individual testing periods on estimation of (co)variance components of a random regression model for feed intake of growing pigs

A simulation study was conducted to assess the influence of differences in the length of individual testing periods on estimates of (co)variance components of a random regression model for daily feed intake of growing pigs performance tested between 30 and 100 kg live weight. A quadratic polynomial in days on test with fixed regressions for sex, random regressions for additive genetic and permanent environmental effects and a constant residual variance was used for a bivariate simulation of feed intake and daily gain. (Co)variance components were estimated for feed intake only by means of a Bayesian analysis using Gibbs sampling and restricted maximum likelihood (REML). A single trait random regression model analogous to the one used for data simulation was used to analyse two versions of the data: full data sets with 18 weekly means of feed intake per animal and reduced data sets with the individual length of testing periods determined when tested animals reached 100 kg live weight. Only one significant difference between estimates from full and reduced data (REML estimate of genetic covariance between linear and quadratic regression parameters) and two significant differences from expected values (Gibbs estimates of permanent environmental variance of quadratic regression parameters) occurred. These differences are believed to be negligible, as the number lies within the expected range of type I error when testing at the 5% level. The course of test day variances calculated from estimates of additive genetic and permanent environmental covariance matrices also supports the conclusion that no bias in estimates of (co)variance components occurs due to the individual length of testing periods of performance‐tested growing pigs. A lower number of records per tested animal only results in more variation among estimates of (co)variance components from reduced compared with full data sets. Compared with the full data, the effective sample size of Gibbs samples from the reduced data decreased to 18% for residual variance and increased up to five times for other (co)variances. The data structure seems to influence the mixing of Gibbs chains.

TRIBES AND CLADES WITHIN APIACEAE SUBFAMILY APIOIDEAE: THE CONTRIBUTION OF MOLECULAR DATA

Phylogenetic analyses of chloroplast gene (rbcL, matK), intron (rpl16, rps16, rpoC1) and nuclear ribosomal DNA internal transcribed spacer (ITS) sequences and chloroplast DNA restriction sites, with supplementary data from variation in size of the chloroplast genome inverted repeat, have been used to elucidate major clades within Apiaceae (Umbelliferae) subfamily Apioideae Drude. This paper summarizes the results of previously published molecular cladistic analyses and presents a provisional classification of the subfamily based on taxonomic congruence among the data sets. Ten tribes (Aciphylleae M. F. Watson & S. R. Downie, Bupleureae Spreng., Careae Baill., Echinophoreae Benth., Heteromorpheae M. F. Watson & S. R. Downie, Oenantheae Dumort., Pleurospermeae M. F. Watson & S. R. Downie, Pyramidoptereae Boiss., Scandiceae Spreng. and Smyrnieae Spreng.) are erected or confirmed as monophyletic, with Scandiceae comprising subtribes Daucinae Dumort., Scandicinae Tausch and Torilidinae Dumort. Seven additional clades are also recognized but have yet to be treated formally, and at least 23 genera examined to date are of dubious tribal or clade placement. The utility of these different molecular markers for phylogenetic inference in Apioideae is compared based on maximum parsimony analyses of subsets of previously published molecular data sets. Of the six loci sequenced, the ITS region is seen to be evolving most rapidly and rbcL is the most conservative. Intermediate in rate of evolution are matK and the three chloroplast introns; with rpl16 and rps16 evolving slightly faster than matK or rpoC1. The analysis of restriction sites, however, provided 2–4 times more parsimony informative characters than any single DNA locus sequenced, with estimates of divergence just slightly lower than that of the ITS region. The trees obtained from separate analyses of these reduced data sets are consistent with regard to the major clades inferred and the relationships among them. Similar phylogenies are obtained by combining data or combining trees, representing the supermatrix and supertree approaches to phylogenetic analysis, respectively. The inferred relationship among the tribes and informally recognized major clades within Apioideae is presented.

Three-dimensional reconstruction of neuronal forests

Three-dimensional reconstruction of neuronal forests from a coherent volumetric data set representing Golgi-stained cortical material is demonstrated. The system automates in parallel neuron feature extraction and reconstruction, replacing largely manual techniques for tracing individual neurons. Serial sections were digitized at 1024×1024 pixels at a linear resolution of 0.37 μm, and then digitally aligned into a coherent volume data set. Our reconstruction program, Recon, determined regions of interest (ROIs) in each image and culled data aggressively, reducing the original volumetric data 70-fold into an ROI-based aligned image stack . The neuronal forest of Fig. 1 below was treaded from this reduced data set.

Analysis of the imputed female urinary incontinence data for the evaluation of expert system parameters

We evaluated parameters for an expert system which will be designed to aid the differential diagnosis of female urinary incontinence by using knowledge discovered from data. To allow the statistical analysis, we applied means, regression and Expectation–Maximization (EM) imputation methods to fill in missing values. In addition, complete-case analysis was performed. Logistic regression results from the imputed data were reasonable. The significant parameters were mostly those that are important in the diagnostic work-up. Moreover, directions of relations between the parameters and the stress, mixed and sensory urge diagnoses were as expected. Analysis with the complete reduced data set gave clearly insufficient results. Imputed values had a moderate agreement, but odds ratios and classification accuracies of logistic regression equations were similar. Results suggest that with these data, simpler methods may be used to allow multivariate analysis and knowledge discovery, when better methods, such as EM imputation, are unavailable. Cluster analysis detected clusters corresponding to the small normal class, but was unable to clearly separate the larger incontinence classes.

Global Event Location with Full and Sparse Data Sets Using Three-dimensional Models of Mantle P-wave Velocity

In order to improve on the accuracy of event locations at teleseismic distances it is necessary to adequately correct for lateral variations in structure along the ray paths, either through deterministic model-based corrections, empirical path/station corrections, or a combination of both approaches. In this paper we investigate the ability of current three-dimensional models of mantle P-wave velocity to accurately locate teleseismic events. We test four recently published models; two are parameterized in terms of relatively long-wavelength spherical harmonic functions up to degree 12, and two are parameterized in terms of blocks of constant velocity which have a dimension of a few hundreds of km. These models, together with detailed crustal corrections, are used to locate a set of 112 global test events, consisting of both earthquakes and explosions with P-wave travel-time data compiled by the International Seismological Centre (ISC). The results indicate that the supposedly higher resolution block models do not improve the accuracy of teleseismic event locations over the longer wavelength spherical harmonic models. For some source locations the block models do not predict the range of observed travel-time residuals as well as the longer wavelength models. The accuracy of the locations largely varies randomly with geographic position although events in central Asia are particularly well located. We also tested the effect of reduced data sets on the locations. Multiple location iterations using 30 P-wave travel times indicate that teleseismic events may be located within an area of 1000 km’ of the true location 66% of the time with only the model-based corrections, and increasing to 75% if calibration information is available. If as few as 8 phases are available then this is possible only 50% of the time. Further refinement in models and/or procedure, such as the addition ofP“phases, azimuth data, and consideration of P-wave anisotropy may provide further improvement in the teleseismic location of small events.

Infraspecific variation in Cerastium alpinum s. lat. (Caryophyllaceae) in Central Norway

WithinCerastium alpinums. lat., several taxa have been circumscribed. Three infraspecific taxa are currently recognized in Nordic literature, but the opinions have varied about their taxonomic levels and relations to each other. Populations of these three taxa —alpinum, glabratumandlanatum— were investigated in Central Norway to elucidate taxonomic relationships on a regional scale. Morphometric analyses (PCA ordination) based on two data sets; one with 61 characters, and one without the 35 indumentum characters, gave different results. The former partly support the use of a higher rank forglabratum, but this analysis may put too much weight on indumentum characters. The reduced data set gave a more continuous transition between the three taxa, with no special status forglabratum.Crossing experiments between the taxa indicated full cross compatibility as far as relative seed set was concerned. Finally, vegetation analysis (DCA ordination) indicated that the three taxa have somewhat different ecological demands. This suggests that they, in spite of hybridization abilities, normally are maintained as separate entities due to habitat differences. A delimitation of three infraspecific taxa at the same level withinC. alpinums. lat. is thus supported by this study. Whether these taxa should be treated as subspecies or varieties strongly depends upon the definitions of these categories. Viewed in a wide geographical context, however, the arguments for subspecific rank are strengthened.

Rheology of sterically stabilized dispersions and latices

Steric stabilization is a method that is often used to properly disperse small particles. It can be applied in aqueous as well as non-aqueous media. The rheological properties of sterically stabilized dispersions are discussed here. The various controlling parameters and the physical mechanisms involved are reviewed. Brownian hard spheres are used as a reference. Scaling relations are presented that make it possible to reduce data sets and to predict properties. Viscosity, yield stress, shear thickening and viscoelasticity are included. The rheological properties are also related to the fundamental colloidal properties of the dispersions under consideration. Quantitative results are available for monodisperse spherical particles, although the effects of particle size distribution can sometimes be predicted also quite well. In other cases the procedures presented here can be used qualitatively to predict viscosities.

Hierachical Clustering For Data Mining By RBF Network

Clustering is one of the dominant techniques of exploratory data analysis. In the context of data mining, one of the powerful clustering methods is the method of orthogonal least squares (OLS) applied to radial basis functions (RBF) network. However, for data mining case, conventional utilization of OLS learning is not desirable even though the learning process feasible for an amount of data at hand. This is due to the fact that some dominant associations in the data can easily obscure many interrelations of interest among the data at hand and also because of this, the rest of the associations for identification can heavily be limited. Therefore, as novel RBF clustering for data mining, the data set is separated into several subsets so that each subset is subjected to clustering separately. The dominant clusters in each subset are accumulated throughout the modular processing of total data set. The newly formed reduced data set which comprises the patterns from the clusters of the subsets, is subjected to final RBF network clustering by OLS for hierarchical cluster gradation of the subsets to identify the existing interrelations being searched for. In this way the associations are discovered in a perspective without obscurity caused by some strong but local associations in the given data set. The paper deals with the details of the novel RBF clustering for data mining.

Automatic detection of buried utilities and solid objects with GPR using neural networks and pattern recognition

The task of locating buried utilities using ground penetrating radar is addressed, and a novel processing technique computationally suitable for on-site imaging is proposed. The developed system comprises a neural network classifier, a pattern recognition stage, and additional pre-processing, feature-extraction and image processing stages. Automatic selection of the areas of the radargram containing useful information results in a reduced data set and hence a reduction in computation time. A backpropagation neural network is employed to identify portions of the radar image corresponding to target reflections by training it to recognise the Welch power spectral density estimate of signal segments reflected from various types of buried target. This results in a classification of the radargram into useful and redundant sections, and further processing is performed only on the former. The Hough Transform is then applied to the edges of these reflections, in order to accurately identify the depth and position of the buried targets. This allows a high resolution reconstruction of the subsurface with reduced computation time. The system was tested on data containing pipes, cables and anti-personnel landmines, and the results indicate that automatic and effective detection and mapping of such structures can be achieved in near real-time.

An objective analysis of the pressure-volume curve in the acute respiratory distress syndrome.

To assess the interobserver and intraobserver variability in the clinical evaluation of the quasi-static pressure-volume (P-V) curve, we analyzed 24 sets of inflation and deflation P-V curves obtained from patients with ARDS. We used a recently described sigmoidal equation to curve-fit the P-V data sets and objectively define the point of maximum compliance increase of the inflation limb (P(mci, i)) and the true inflection point of the deflation limb (P(inf,d)). These points were compared with graphic determinations of lower Pflex by seven clinicians. The graphic and curve-fitting methods were also compared for their ability to reproduce the same parameter value in data sets with reduced number of data points. The sigmoidal equation fit the P-V data with great accuracy (R(2) = 0.9992). The average of Pflex determinations was found to be correlated with P(mci,i) (R = 0.89) and P(inf,d) (R = 0.76). Individual determinations of Pflex were less correlated with the corresponding objective parameters (R = 0.67 and 0.62, respectively). Pflex + 2 cm H(2)O was a more accurate estimator of P(inf,d) (2 SD = +/-6.05 cm H(2)O) than Pflex was of P(mci,i) (2 SD = +/-8.02 cm H(2)O). There was significant interobserver variability in Pflex, with a maximum difference of 11 cm H(2)O for the same patient (SD = 1.9 cm H(2)O). Clinicians had difficulty reproducing Pflex in smaller data sets with differences as great as 17 cm H(2)O (SD = 2.8 cm H(2)O). In contrast, the curve-fitting method reproduced P(mci,i) with great accuracy in reduced data sets (maximum difference of 1.5 cm H(2)O and SD = 0.3 cm H(2)O). We conclude that Pflex rarely coincided with the point of maximum compliance increase defined by a sigmoid curve-fit with large differences in Pflex seen both among and within observers. Calculating objective parameters such as P(mci,i) or P(inf,d) from curve-fitted P-V data can minimize this large variability.

Computational electrochemistry: aqueous one-electron oxidation potentials for substituted anilines

Semiempirical molecular orbital theory and density functional theory are used to compute one-electron oxidation potentials for aniline and a set of 21 mono- and di-substituted anilines in aqueous solution. Linear relationships between theoretical predictions and experiment are constructed and provide mean unsigned errors as low as 0.02 V over a training set of 13 anilines; the error rises to 0.09 V over a test set of eight additional anilines. A good correlation is also found between oxidation potential and a simple computed property, namely the energy of the highest occupied molecular orbital for neutral anilines in aqueous solution. For the particular case of the substituted anilines, a strong correlation between oxidation potential and pKa is found, and a still stronger correlation between oxidation potential and physical organic descriptors for aromatic substituents is also found, albeit over a reduced data set.

Accuracy of quantitative ventriculographic analysis at reduced frame rates.

We measured the error in left ventricular (LV) volume and wall motion measurement from recording contrast ventriculograms at 15 frames/sec instead of 30 frames/sec. Angiograms are increasingly recorded at 15 frames/sec instead of 30 frames/sec to improve efficiency. However the resulting error in quantitative analysis has not been evaluated. Contrast ventriculograms recorded at 30 frames/sec of 23 patients undergoing diagnostic studies were digitized from cine film. End diastolic volume (EDV), end systolic volume (ESV), ejection fraction, anterior wall motion, and inferior wall motion were measured from a full set recorded at 30 frames/sec and from two reduced data sets at 15 frames/sec generated from the even and odd numbered frames. Intraobserver variability was compared with the error due to frame rate reduction. The error due to frame rate reduction did not differ significantly from intraobserver variability (p > 0.1 for all), measuring 2.6 +/- 2.2 ml (r = 0.997) vs. 1.9 +/- 1.2 ml (r = 0.999), respectively for EDV, 2.4 +/- 1.8 ml (r = 0.991) vs. 2.1 +/- 1.7 ml (r = 0.994), respectively for ESV, and 0.2 +/- 0.2 percent (r = 0. 981) vs. 0.1 +/- 1.2 percent (r = 0.994), respectively for ejection fraction. The error and variability for wall motion were 0.3 +/- 0.3 SD (r = 0.928) vs. 0.2 +/- 0.1 SD (r = 0.946), respectively for the anterior wall and 0.2 +/- 0.2 SD (r = 0.963) vs. 0.2 +/- 0.1 SD (r = 0.968), respectively for the inferior wall. Imaging the LV at 15 frames/sec does not cause a significant error in measuring LV volume and function compared with imaging at 30 frames/sec. Cathet. Cardiovasc, Intervent. 48:18-21, 1999.

A proposal for a standard electronic anthropomorphic phantom for radiotherapy.

To create a database of human male and female computed tomography (CT) slices, the National Library of Medicine organized the "Visible Human Project." Since the male and female data sets provided are approximately 269 MB and 915 MB, respectively, both the size and complexity have been reduced. While making the slices accessible to those with limited computing resources, the production of these reduced data sets also presents a unique opportunity to establish a standard human CT slice library for research in radiation therapy. A brief history of the original data sets is included, as well as details of the reduction process, applications to radiotherapy, and information on accessing these reduced image files.

Palaeointensity results from Ethiopian basalts: implications for the Oligocene geomagnetic field strength

From a large collection of Ethiopian flood basalts (~30 Myr old) sampled for magnetostratigraphy, 40Ar/39Ar geochronology and geochemical investigations, 47 samples were selected in order to test their suitability for Thellier palaeointensity experiments. Only 17 samples from eight individual flows yielded reliable palaeointensity estimates, with flow-mean virtual dipole moments ranging from 3.0 to 10.5×1022 Am2. A critical review of the Oligocene palaeointensity data set, including these new Ethiopian data, indicates an Oligocene mean virtual dipole moment of 5.1±2.5×1022 Am2 for the complete data set. After applying mild selection criteria, the reduced data set yields a mean value of only 4.6±1.9×1022 Am2. This value is significantly lower than the present-day field strength but is higher than the Mesozoic dipole low mean field. This low Oligocene field might be in agreement with the high palaeosecular variation and rather high non-dipole field invoked around 30 Ma. However, the Oligocene data set is largely dependent on the palaeointensity determinations from Armenia, obtained mainly from baked contacts, which show amazingly low dispersion at both flow and between-flow levels. More data are needed to reduce the weight of these determinations on the mean value and avoid a possible bias.

Comparison of three different QSAR/QSPR generation techniques

C.M. Breneman and M. Rhem [Journal of Computational Chemistry,18 (1997) 182], recently published data sets containing electronic van der Waals surface property descriptors and the experimental HPLC capacity factors using either ODS or CPK columns for a series of compounds. This study uses one of these data sets (the HPLC ODS column capacity factors [log( k′)]) in a Quantitative Structure/Property analysis. The chosen data set contains 22 molecules with 118 descriptors each. Two simple prescreening methods are examined to determine if they can properly reduce the number of descriptors. One method is found to fail, while the other is used to reduce the number of descriptors to 111. Three different computational methods are used to generate 5-descriptor, linear relationships with this reduced data set. The first, and fastest method uses a series of 1-dimensional fits to the capacity factors and residual errors. The second method uses an Evolutionary Programming (EP) paradigm to choose all five descriptors and the third is a complete search over all sets of five descriptors. This paper outlines each of these computational methods and presents their accuracy and CPU requirements. Though the EP method does outperform the stepwise fit, it initially fails to find the best relationships. A close examination of the terms present in the 100 best relationships shows why this data set is not amenable to a QSPR generation technique that uses either a Genetic Algorithm or Evolutionary Programming. By allowing the EP program to retain relationships with an incorrect number of descriptors, it is able to find the best overall 5-descriptor, linear relationship.

Plant communities of New Brunswick in relation to environmental variation

Abstract. The objective of this study was to quantitatively describe vegetation‐environment relationships at a regional scale within the Province of New Brunswick, Canada, using vegetation and environment data from 3947 provincial forestry sample plots. The major plant community types in the province were identified using cluster analysis. Relationships of these communities to climate, topography and soil variables were analyzed by Canonical Correspondence Analysis (CCA), using both a reduced data set consisting of cluster likelihood scores × sample plots and an unreduced species × sample plots data matrix. The vegetation types and major axes of environmental variation were mapped to examine the geographic distributions of these factors within the province. Eight communities were identified and described in terms of enhanced/reduced species (significantly higher or lower frequencies of occurrence in a specific community type relative to all plots) and common species (species in the community type with the highest frequencies of occurrence). The canonical axes explained 25 % of the variation in the vegetation cluster data. Vegetation composition was related to three major environmental gradients representing climate and elevation, soil moisture, and soil fertility. The geographic distributions of vegetation communities exhibited predictable but weak correspondence to the geographic distributions of individual environmental factors. Our findings emphasize the overriding importance of climate and topography and the secondary importance of soil factors in controlling vegetation pattern at the regional scale.

Distribution and floristics of moss- and lichen-dominated soil crusts in a patterned Callitris glaucophylla woodland in eastern Australia

The distribution and abundance of soil crust lichens and bryophytes was examined in a patterned Callitris glaucophylla woodland in eastern Australia. Twenty-one lichen species and 26 bryophyte species were collected within thirty quadrats along a sequence of runoff, interception and runoff zones. Crust cover was significantly greatest in the interception zones (79.0 %), followed by the runoff zones (24.0 %), and lowest in the groved, runon zones (6.6 %). Lichens and bryophytes were distributed across all geomorphic zones, and, although there were significantly more moss species in the interception zones (mean = 9.1) compared with either the runoff (4.2) or runon (3.2) zones, the number of lichen species did not vary between zones. Ordination of a reduced data set of 32 species revealed a separation of taxa into distinct groups corresponding to the three geomorphic zones. Canonical correspondence analysis (CCA) of the 32 species and thirteen environmental variables revealed that the most important factors associated with the distribution of species were sheet and scarp erosion, soil stability and coherence, litter cover and crust cover. Surface cracking, microtopography and plant cover were of intermediate importance. The CCA biplot revealed that the timbered runon zones (groves) were dominated by `shade-tolerant' mosses Fissidens vittatus and Barbula hornschuchiana, whilst the heavily eroded runoff zones supported sparse populations of `erosion tolerant' lichens (Endocarpon rogersii) and mosses (Bryum argenteum and Didymodon torquatus). Interception zones supported a rich suite of `crust forming' mosses and lichens capable of tolerating moderate inundation by overland flow. Two other groups of taxa were identified by this analysis: the `pioneer' group, comprising mainly nitrogen-fixing lichens which occupy the zone of active erosion at the lower edge of the groves, and the `opportunists' dominated by liverworts, occupying the shallow depressions or bays at the margins of the groves and the interception zones. This study confirms that the non-vascular lichens and bryophytes in these arid soil crusts, are, like the vascular plants, strongly patterned according to geomorphic zone, being most strongly associated with soil surface and erosional features.

Determining protein structure using the distance geometry program APA

APA is a computer program, written in C, designed to determine the three-dimensional structure of proteins using distance geometry. We present the sampling and convergence properties of APA, as tested on bovine pancreatic trypsin inhibitor (BPTI). The results confirm the program’s earlier success with poly- l-alanine, albeit with some complications. The correct overall orientation of the BPTI conformation is achieved at an early stage in the algorithm. The correct orientations of the α-carbons are obtained by local reflections, instead of a penalty term, resulting in a smoother convergence. Finally, a process of choosing dissimilarities from two reduced data sets resulted in almost all structures converging. In order to compare with Havel’s DG-II distance geometry program, the sampling and convergence properties were tested on Havel’s 10 data sets. These simulated data sets were generated from the BPTI crystal and kindly provided by Tim Havel.

FAST MOTION ESTIMATION SCHEME FOR VIDEO CODING USING FEATURE VECTOR MATCHING AND MOTION VECTOR'S CORRELATIONS

In this paper, we propose a fast block motion estimation scheme by exploiting the correlations of motion vectors (MV) existing in spatially- and temporally-adjacent as well as hierarchically-related blocks. The basic idea is to use the information obtained from the corresponding block at a coarser resolution level and spatio-temporal neighboring blocks at the same level in order to select a good set of initial MV candidates and then perform further local search to find the best matching MV. In order to further reduce computational complexity, the sign pattern vector (SPV) is defined and used for block matching, as opposed to the pixel intensity values used in the conventional block matching methods. By using the SPV definition, a data block can be presented by a mean and a set of binary sign patterns with a much reduced data set. The block matching motion estimation is then divided into mean matching and binary phase matching. The proposed technique enables a significant reduction in computational complexity compared with the conventional block matching motion estimation (ME) because binary phase matching only involves Boolean logic operations. This scheme also significantly reduces the data transfer time between the frame buffer and motion estimator. Our test results indicate that the performance of the proposed scheme is comparable with the full-search block matching under the same search ranges, however, the proposed scheme has a speed-up factor ranging from 250 to 370 in comparison with full search.