Aggregation functions based on penalties

This article studies a large class of averaging aggregation functions based on minimizing a distance from the vector of inputs, or equivalently, minimizing a penalty imposed for deviations of individual inputs from the aggregated value. We provide a systematization of various types of penalty based aggregation functions, and show how many special cases arise as the result. We show how new aggregation functions can be constructed either analytically or numerically and provide many examples. We establish connection with the maximum likelihood principle, and present tools for averaging experimental noisy data with distinct noise distributions.

OWA operators in linear regression and detection of outliers

We consider the use of Ordered Weighted Averaging (OWA) in linear regression. Our goal is to replace the traditional least squares, least absolute deviation, and maximum likelihood criteria with an OWA function of the residuals. We obtain several high breakdown robust regression methods as special cases (least median, least trimmed squares, trimmed likelihood methods). We also present new formulations of regression problem. OWA-based regression is particularly useful in the presence of outliers.

OWA operators in regression problems

We consider an application of fuzzy logic connectives to statistical regression. We replace the standard least squares, least absolute deviation, and maximum likelihood criteria with an ordered weighted averaging (OWA) function of the residuals. Depending on the choice of the weights, we obtain the standard regression problems, high-breakdown robust methods (least median, least trimmed squares, and trimmed likelihood methods), as well as new formulations. We present various approaches to numerical solution of such regression problems. OWA-based regression is particularly useful in the presence of outliers, and we illustrate the performance of the new methods on several instances of linear regression problems with multiple outliers.

Bearing-only localization using geometrically constrained optimization

We examine the problem of optimal bearing-only localization of a single target using synchronous measurements from multiple sensors. We approach the problem by forming geometric relationships between the measured parameters and their corresponding errors in the relevant emitter localization scenarios. Specifically, we derive a geometric constraint equation on the measurement errors in such a scenario. Using this constraint, we formulate the localization task as a constrained optimization problem that can be performed on the measurements in order to provide the optimal values such that the solution is consistent with the underlying geometry. We illustrate and confirm the advantages of our approach through simulation, offering detailed comparison with traditional maximum likelihood (TML) estimation.

Comparison of automated classification techniques for predicting benthic biological communities using hydroacoustics and video observations

The effective management of our marine ecosystems requires the capability to identify, characterise and predict the distribution of benthic biological communities within the overall seascape architecture. The rapid expansion of seabed mapping studies has seen an increase in the application of automated classification techniques to efficiently map benthic habitats, and the need of techniques to assess confidence of model outputs. We use towed video observations and 11 seafloor complexity variables derived from multibeam echosounder (MBES) bathymetry and backscatter to predict the distribution of 8 dominant benthic biological communities in a 54 km² site, off the central coast of Victoria, Australia. The same training and evaluation datasets were used to compare the accuracies of a Maximum Likelihood Classifier (MLC) and two new generation decision tree methods, QUEST (Quick Unbiased Efficient Statistical Tree) and CRUISE (Classification Rule with Unbiased Interaction Selection and Estimation), for predicting dominant biological communities. The QUEST classifier produced significantly better results than CRUISE and MLC model runs, with an overall accuracy of 80% (Kappa 0.75). We found that the level of accuracy with the size of training set varies for different algorithms. The QUEST results generally increased in a linear fashion, CRUISE performed well with smaller training data sets, and MLC performed least favourably overall, generating anomalous results with changes to training size. We also demonstrate how predicted habitat maps can provide insights into habitat spatial complexity on the continental shelf. Significant variation between patch-size and habitat types and significant correlations between patch size and depth were also observed.

Range size-abundance relationships in Australian passerines

Aim  To investigate the relationship between geographical range size and abundance (population density) in Australian passerines.
Location  Australia (including Tasmania).
Methods   We analysed the relationship between range size and local abundance for 272 species of Australian passerines, across the whole order and within families. We measured abundance as mean and maximum abundance, and used a phylogenetic generalized least-squares regression method within a maximum-likelihood framework to control for effects of phylogeny. We also analysed the relationship within seven different habitat types.
Results  There was no correlation between range size and abundance for the whole set of species across all habitats. Analyses within families revealed some strong correlations but showed no consistent pattern. Likewise we found little evidence for any relationship or conflicting patterns in different habitats, except that woodland/forest habitat species exhibit a negative correlation between mean abundance and range size, whilst species in urban habitats exhibit a significant positive relationship between maximum abundance and range size. Despite the general lack of correlation, the raw data plots of range size and abundance in this study occupied a triangular space, with narrowly distributed species exhibiting a greater variation in abundances than widely distributed species. However, using a null model analysis, we demonstrate that this was due to a statistical artefact generated by the frequency distributions for the individual variables.
Conclusions   We find no evidence for a positive range size-abundance relationship among Australian passerines. This absence of a relationship cannot be explained by any conflicting effects introduced by comparing across different habitats, nor is it explained by the fact that large proportions of Australia are arid. We speculate that the considerable isolation and evolutionary age of Australian passerines may be an explanatory factor.

Software reliability growth models based on local polynomial modeling with kernel smoothing

Software reliability growth models (SRGMs) are extensively employed in software engineering to assess the reliability of software before their release for operational use. These models are usually parametric functions obtained by statistically fitting parametric curves, using Maximum Likelihood estimation or Least–squared method, to the plots of the cumulative number of failures observed N(t) against a period of systematic testing time t. Since the 1970s, a very large number of SRGMs have been proposed in the reliability and software engineering literature and these are often very complex, reflecting the involved testing regime that often took place during the software development process. In this paper we extend some of our previous work by adopting a nonparametric approach to SRGM modeling based on local polynomial modeling with kernel smoothing. These models require very few assumptions, thereby facilitating the estimation process and also rendering them more relevant under a wide variety of situations. Finally, we provide numerical examples where these models will be evaluated and compared.

Blood mineral, trace-element and vitamin concentrations in Huacaya alpacas and Merino sheep grazing the same pasture

We aimed to determine whether the concentration of minerals and trace constituents in blood of Merino sheep and Huacaya alpacas grazing the same pasture differed with species and time of sampling. Blood samples and pasture samples were collected at frequent intervals over a period of 2 years for mineral and trace-nutrient assay. The concentration of the minerals and trace nutrients in the grazed pasture usually met the dietary needs of sheep at maintenance, apart from potassium, sulfur, cobalt and Vitamin E in occasional samples. Restricted maximum likelihood mixed model analysis indicated a significant (P < 0.001) species by month by year interaction for all blood constituents assayed, a significant (P < 0.05) species by coat shade interaction for plasma Vitamin D, E and B12 and a significant (P < 0.001) species by month by Vitamin D interaction for plasma phosphorus concentrations. In general, plasma calcium concentrations were greater in sheep than in alpacas but plasma magnesium concentrations were greater in alpacas than in sheep. There was no consistent difference between the two species in plasma phosphorus concentrations although low values were recorded in individual sheep and alpacas. Plasma Vitamin D concentrations were more responsive to increasing hours of sunlight in alpacas than they were in sheep. Sheep had consistently higher concentrations of plasma copper, zinc and Vitamin B12 and higher concentrations of blood selenium but lower concentrations of plasma selenium and Vitamin A, than did alpacas. No consistent difference was observed between the two species in plasma Vitamin E concentrations.

The hidden permutation model and location-based activity recognition

Permutation modeling is challenging because of the combinatorial nature of the problem. However, such modeling is often required in many real-world applications, including activity recognition where subactivities are often permuted and partially ordered. This paper introduces a novel Hidden Permutation Model (HPM) that can learn the partial ordering constraints in permuted state sequences. The HPM is parameterized as an exponential family distribution and is flexible so that it can encode constraints via different feature functions. A chain-flipping Metropolis-Hastings Markov chain Monte Carlo (MCMC) is employed for inference to overcome the O(n!) complexity. Gradient-based maximum likelihood parameter learning is presented for two cases when the permutation is known and when it is hidden. The HPM is evaluated using both simulated and real data from a location-based activity recognition domain. Experimental results indicate that the HPM performs far better than other baseline models, including the naive Bayes classifier, the HMM classifier, and Kirshner's multinomial permutation model. Our presented HPM is generic and can potentially be utilized in any problem where the modeling of permuted states from noisy data is needed.

AdaBoost.MRF: boosted Markov random forests and application to multilevel activity recognition

Activity recognition is an important issue in building intelligent monitoring systems. We address the recognition of multilevel activities in this paper via a conditional Markov random field (MRF), known as the dynamic conditional random field (DCRF). Parameter estimation in general MRFs using maximum likelihood is known to be computationally challenging (except for extreme cases), and thus we propose an efficient boosting-based algorithm AdaBoost.MRF for this task. Distinct from most existing work, our algorithm can handle hidden variables (missing labels) and is particularly attractive for smarthouse domains where reliable labels are often sparsely observed. Furthermore, our method works exclusively on trees and thus is guaranteed to converge. We apply the AdaBoost.MRF algorithm to a home video surveillance application and demonstrate its efficacy.

The effect of humidity and temperature on Wool Comfort Meter assessment of single jersey wool fabrics

The Wool ComfortMeter provides an objective measurement of the fabric-evoked prickle discomfort rating provided by wearers. This work aimed to quantify the sensitivity of the Wool ComfortMeter over a range of different temperature and humidity conditions to determine the recommended test conditions for its operation. The design was: three temperatures (notionally 20, 25 and 30°C) at three relative humidities (RHs, notionally 50, 65 and 80%) each with two replicates, using six different wool single jersey knits (mean fibre diameter 19.5–27.0 µm). As it was difficult to achieve exactly some of the extreme combinations of temperature and RH, some combinations were repeated, providing a total of 23 different assessment conditions. Data were analysed using restricted maximum likelihood mixed model analysis. The best fixed model included RH, RH2, temperature and the interaction of temperature and RH, accounting for 95% of the variation in Wool ComfortMeter readings. Wool ComfortMeter values were almost constant at 55–60% RH. Generally, the Wool ComfortMeter value reduced with increasing RH > 60% at temperatures of 25°C and 28.5°C as the regain of the fabric increased. However, at 20°C little change was detected as RH was increased from 50 to 80% as there were only small changes in fabric regain. The observed effects were in a good agreement with existing knowledge on the effect of regain on the mechanical properties of wool fibre. Wool ComfortMeter is best operated under standard conditions for textile testing of 65% RH and 20°C.

Production, properties and processing of American bison (Bison bison) wool grown in southern Australia

American bison grow a thick coat of fibres which assists them to withstand severe climatic conditions. Bison fibre was traditionally used in textiles by native North Americans. This study aimed to quantify the production, fibre attributes and dehairing processing of bison fibre produced from bison grazed in north-eastern Victoria. Three age/sex classes were sampled (n = 16) at seven body positions in spring. The fibre growing area was measured. Fibre was tested for diameter distribution, clean washing yield, proportion of fine fibres <36µm and fine fibre length, and processed by cashmere dehairing. Bison were 12 years of age, liveweights 160450 kg and had mean fibre growing area of 1.4 m². They produced an average 1184 g (range 5301640 g) of fine fibre with mean fibre diameter 18.5µm, clean washing yield 76.5%, wax content 9.8%, suint content 14.5%, clean fine fibre yield 56.4%, fine fibre length 37 mm and fibre curvature was 93/mm. Mid-side fibre had a crimp frequency of 6.5/cm and mean resistance to compression of 6.6 kPa. Fibre had a tenacity of 8.7 cN/tex and an extension of 39.3%. Restricted maximum likelihood mixed model analysis showed age/sex class and sampling site significantly affected all fibre attributes. Finer and longer fibre was produced in anterior sites and in younger bison. Fibre curvature declined 5.3°/mm for each 1-µm increase in mean fibre diameter. Dehaired fibre had a mean fibre diameter of 17.8 µm and mid-length of 28 mm, suitable for woollen spinning. The production by bison of coats containing significant amounts of fibre indicates that careful harvesting of fibre could form an important source of income in bison enterprises.

The allometric relationship between mean fibre diameter of mohair and the fleece-free liveweight of Angora goats over their lifetime

As mean fibre diameter (MFD) is the primary determinant of mohair price we aimed to quantify the lifetime changes in mohair MFDas Angora goats aged and grew. Measurements were made over 12 shearing periods on a population of Angora goats representing the current range and diversity of genetic origins including South African, Texan and interbred admixtures of these and Australian sources. Records of sire, dam, birthweight, birth parity, liveweight, fleece growth and fleece quality were taken for does and castrated males (wethers) (n = 267 animals). Fleece-free liveweights (FFL_wt) were determined for each goat at shearing time by subtracting the greasy fleece weight from the liveweight recorded immediately before shearing. A restricted maximum likelihood growth curve model was developed for relating MFD to FFL_wt, age and other measurements.Asimple way of describing the results is:MFD= k (FFL_wt)b E; where k is a parameter that can vary in a systematic way with shearing(age), breed, weaning weight, sire, dam and individual; b is a parameter that is the same for nearly the whole study; and E are independent errors from a log-normal distribution. The analysis shows that ^b = 0.34, with s.e. (^b) = 0.021. Thus, mohair MFD was allometrically related to the cube root of FFLwt over the lifetime of Angora goats. However, the allometric proportionality constant differed in a systematic way with age at shearing, genetic strain, weaning weight, sire, dam and individual. For Texan-breed goats, MFD decreased as weaning weight increased (P = 0.00016). The findings indicate that management factors that affect liveweight and weaning weight have lifetime effects on mohair fibre diameter and therefore the value of mohair and the profitability of the mohair enterprise.

The allometric relationship between clean mohair growth and the fleece-free liveweight of Angora goats is affected by liveweight change

Clean fleece weight (CFWt) is affected by liveweight and change in liveweight in Merino sheep, Angora and cashmere goats. However, how these relationships progress as animals age has not been elucidated. Measurements were made over 12 shearing periods on a population of Angora goats representing the current range and diversity of genetic origins including South African, Texan and interbred admixtures of these and Australian sources. Records of breed, sire, dam, date of birth, dam age, birthweight, birth parity, weaning weight, liveweight, fleece growth and fleece quality were taken for does and castrated males (wethers) (n = 267 animals). Fleece-free liveweights (FFLwt) were determined for each goat at shearing time by subtracting the greasy fleece weight from the liveweight recorded immediately before shearing. The average of the FFLwt at the start of the period and the FFLWt at the end of the period was calculated (AvFFLwt). Liveweight change (LwtCh) was the change in FFLwt over the period between shearings. A restricted maximum likelihood model was developed for CFWt, after log10 transformation, which allowed the observations of the same animal at different ages to be correlated in an unstructured manner. A simple way of describing the results is: CFWt = κ (AvFFLwt)β, where κ is a parameter that can vary in a systematic way with shearing age, shearing treatment and LwtCh; and β is an allometric coefficient that only varies with LwtCh. CFWt was proportional to FFLwt0.67 but only when liveweight was lost at the rate of 5–10 kg during a shearing interval of 6 months. The allometric coefficient declined to 0.3 as LwtCh increased from 10 kg loss to 20 kg gain during a shearing interval. A consequence is that, within an age group of Angora goats, the largest animals will be the least efficient in converting improved nutrition to mohair.

Evaluation of four supervised learning methods for benthic habitat mapping using backscatter from multi-beam sonar

An understanding of the distribution and extent of marine habitats is essential for the implementation of ecosystem-based management strategies. Historically this had been difficult in marine environments until the advancement of acoustic sensors. This study demonstrates the applicability of supervised learning techniques for benthic habitat characterization using angular backscatter response data. With the advancement of multibeam echo-sounder (MBES) technology, full coverage datasets of physical structure over vast regions of the seafloor are now achievable. Supervised learning methods typically applied to terrestrial remote sensing provide a cost-effective approach for habitat characterization in marine systems. However the comparison of the relative performance of different classifiers using acoustic data is limited. Characterization of acoustic backscatter data from MBES using four different supervised learning methods to generate benthic habitat maps is presented. Maximum Likelihood Classifier (MLC), Quick, Unbiased, Efficient Statistical Tree (QUEST), Random Forest (RF) and Support Vector Machine (SVM) were evaluated to classify angular backscatter response into habitat classes using training data acquired from underwater video observations. Results for biota classifications indicated that SVM and RF produced the highest accuracies, followed by QUEST and MLC, respectively. The most important backscatter data were from the moderate incidence angles between 30° and 50°. This study presents initial results for understanding how acoustic backscatter from MBES can be optimized for the characterization of marine benthic biological habitats. © 2012 by the authors.