Robust face clustering for real-world data

This thesis has investigated how to cluster a large number of faces within a multi-media corpus in the presence of large session variation. Quality metrics are used to select the best faces to represent a sequence of faces; and session variation modelling improves clustering performance in the presence of wide variations across videos. Findings from this thesis contribute to improving the performance of both face verification systems and the fully automated clustering of faces from a large video corpus.

Global, regional, and national incidence, prevalence, and years lived with disability for 301 acute and chronic diseases and injuries in 188 countries, 1990–2013: a systematic analysis for the Global Burden of Disease Study 2013

Background Up-to-date evidence about levels and trends in disease and injury incidence, prevalence, and years lived with disability (YLDs) is an essential input into global, regional, and national health policies. In the Global Burden of Disease Study 2013 (GBD 2013), we estimated these quantities for acute and chronic diseases and injuries for 188 countries between 1990 and 2013. Methods Estimates were calculated for disease and injury incidence, prevalence, and YLDs using GBD 2010 methods with some important refi nements. Results for incidence of acute disorders and prevalence of chronic disorders are new additions to the analysis. Key improvements include expansion to the cause and sequelae list, updated systematic reviews, use of detailed injury codes, improvements to the Bayesian meta-regression method (DisMod-MR), and use of severity splits for various causes. An index of data representativeness, showing data availability, was calculated for each cause and impairment during three periods globally and at the country level for 2013. In total, 35 620 distinct sources of data were used and documented to calculated estimates for 301 diseases and injuries and 2337 sequelae. The comorbidity simulation provides estimates for the number of sequelae, concurrently, by individuals by country, year, age, and sex. Disability weights were updated with the addition of new population-based survey data from four countries. Findings Disease and injury were highly prevalent; only a small fraction of individuals had no sequelae. Comorbidity rose substantially with age and in absolute terms from 1990 to 2013. Incidence of acute sequelae were predominantly infectious diseases and short-term injuries, with over 2 billion cases of upper respiratory infections and diarrhoeal disease episodes in 2013, with the notable exception of tooth pain due to permanent caries with more than 200 million incident cases in 2013. Conversely, leading chronic sequelae were largely attributable to non-communicable diseases, with prevalence estimates for asymptomatic permanent caries and tension-type headache of 2∙4 billion and 1∙6 billion, respectively. The distribution of the number of sequelae in populations varied widely across regions, with an expected relation between age and disease prevalence. YLDs for both sexes increased from 537∙6 million in 1990 to 764∙8 million in 2013 due to population growth and ageing, whereas the age-standardised rate decreased little from 114∙87 per 1000 people to 110∙31 per 1000 people between 1990 and 2013. Leading causes of YLDs included low back pain and major depressive disorder among the top ten causes of YLDs in every country. YLD rates per person, by major cause groups, indicated the main drivers of increases were due to musculoskeletal, mental, and substance use disorders, neurological disorders, and chronic respiratory diseases; however HIV/AIDS was a notable driver of increasing YLDs in sub-Saharan Africa. Also, the proportion of disability-adjusted life years due to YLDs increased globally from 21·1% in 1990 to 31·2% in 2013. Interpretation Ageing of the world’s population is leading to a substantial increase in the numbers of individuals with sequelae of diseases and injuries. Rates of YLDs are declining much more slowly than mortality rates. The non-fatal dimensions of disease and injury will require more and more attention from health systems. The transition to nonfatal outcomes as the dominant source of burden of disease is occurring rapidly outside of sub-Saharan Africa. Our results can guide future health initiatives through examination of epidemiological trends and a better understanding of variation across countries.

Characteristics of on-road driving by persons with central vision loss: Learning to drive with a bioptic telescope

There is limited research on the driving performance and safety of bioptic drivers and even less regarding the driving skills that are most challenging for those learning to drive with bioptic telescopes. This research consisted of case studies of five trainee bioptic drivers whose driving skills were compared with those of a group of licensed bioptic drivers (n = 23) while they drove along city, suburban, and controlled-access highways in an instrumented dual-brake vehicle. A certified driver rehabilitation specialist was positioned in the front passenger seat to monitor safety and two backseat evaluators independently rated driving using a standardized scoring system. Other aspects of performance were assessed through vehicle instrumentation and video recordings. Results demonstrate that while sign recognition, lane keeping, steering steadiness, gap judgments and speed choices were significantly worse in trainees, some driving behaviors and skills, including pedestrian detection and traffic light recognition were not significantly different to those of the licensed drivers. These data provide useful insights into the skill challenges encountered by a small sample of trainee bioptic drivers which, while not generalizable because of the small sample size, provide valuable insights beyond that of previous studies and can be used as a basis to guide training strategies.

Body composition and components of energy expenditure in children with end-stage liver disease

Background: Better understanding of body composition and energy metabolism in pediatric liver disease may provide a scientific basis for improved medical therapy aimed at achieving optimal nutrition, slowing progression to end-stage liver disease (ESLD), and improving the outcome of liver transplantation. Methods: Twenty-one children less than 2 years of age with ESLD awaiting liver transplantation and 15 healthy, aged-matched controls had body compartment analysis using a four compartment model (body cell mass, fat mass, extracellular water, and extracellular solids). Subjects also had measurements of resting energy expenditure (REE) and respiratory quotient (RQ) by indirect calorimetry. Nine patients and 15 control subjects also had measurements of total energy expenditure (TEE) using doubly labelled water. Results: Mean weights and heights were similar in the two groups. Compared with control subjects, children with ESLD had higher relative mean body cell mass (33 ± 2% vs 29 ± 1% of body weight, P < 0.05), but had similar fat mass, extracellular water, and extracellular solid compartments (18% vs 20%, 41% vs 38%, and 7% vs 13% of body weight respectively). Compared with control subjects, children with ESLD had 27% higher mean REE/body weight (0.285 ± 0.013 vs 0.218. ± 0.013 mJ/kg/24h, P < 0.001), 16% higher REE/unit cell mass (P < 0.05); and lower mean RQ (P < 0.05). Mean TEE of patients was 4.70 ± 0.49 mJ/24h vs 3.19 ± 0.76 in controls, (P < 0.01). Conclusions: In children, ESLD is a hypermetabolic state adversely affecting the relationship between metabolic and non-metabolic body compartments. There is increased metabolic activity within the body cell mass with excess lipid oxidation during fasting and at rest. These findings have implications for the design of appropriate nutritional therapy.

The effect of transforming growth factor β1 gene polymorphisms in ankylosing spondylitis

Objectives. To determine whether genetic polymorphisms in or near the transforming growth factor β1 (TGFB1) locus were associated d with susceptibility to or severity of ankylosing spondylitis (AS). Methods. Five intragenic single-nucleotide polymorphisms (SNP) and three microsatellite markers flanking the TGFB1 locus were genotyped. Seven hundred and sixty-two individuals from 184 multiplex families were genotyped for the microsatellite markers and two of the promoter SNPs. One thousand and two individuals from 212 English and 170 Finnish families with AS were genotyped for all five intragenic SNPs. A structured questionnaire was used to assess the age of symptom onset, disease duration and disease severity scores, including the BASDAI (Bath Ankylosing Spondylitis Disease Activity Index) and BASFI (Bath Ankylosing Spondylitis Functional Index). Results. A weak association was noted between the rare TGFB1 + 1632 T allele and AS in the Finnish population (P = 0.04) and in the combined data set (P = 0.03). No association was noted between any other SNPs or SNP haplotype and AS, even among those families with positive non-parametric linkage scores. The TGFB1 +1632 polymorphism was also associated with a younger age of symptom onset (English population, allele 2 associated with age of onset greater by 4.2 yr, P = 0.05; combined data set, allele 2 associated with age of onset greater by 3.2 yr, P = 0.02). A haplotype of coding region SNPs (TGFB1 +869/ +915+1632 alleles 2/1/2) was associated with age of symptom onset in both the English parent-case trios and the combined data set (English data set, haplotype 2/1/2 associated with age of onset greater by 4.9 yr, P = 0.03; combined data set, haplotype 2/1/2 associated with greater age of onset by 4.2 yr, P = 0.006). Weak linkage with AS susceptibility was noted and the peak LOD score was 1.3 at distance 2 cM centromeric to the TGFB1 gene. No other linkage or association was found between quantitative traits and the markers. Conclusion. This study suggests that the polymorphisms within the TGFB1 gene play at most a small role in AS and that other genes encoded on chromosome 19 are involved in susceptibility to the disease.

Site and gender specificity of inheritance of bone mineral density

Differences in genetic control of BMD by skeletal sites and genders were examined by complex segregation analysis in 816 members of 147 families with probands with extreme low BMD. Spine BMD correlated more strongly in male-male comparisons and hip BMD in female-female comparisons, consistent with gender- and site-specificity of BMD heritability. Introduction: Evidence from studies in animals and humans suggests that the genetic control of bone mineral density (BMD) may differ at different skeletal sites and between genders. This question has important implications for the design and interpretation of genetic studies of osteoporosis. Methods: We examined the genetic profile of 147 families with 816 individuals recruited through probands with extreme low BMD (T-score < −2.5, Z-score < −2.0). Complex segregation analysis was performed using the Pedigree Analysis Package. BMD was measured by DXA at both lumbar spine (L1-L4) and femoral neck. Results: Complex segregation analysis excluded purely monogenic and environmental models of segregation of lumbar spine and femoral neck BMD in these families. Pure polygenic models were excluded at the lumbar spine when menopausal status was considered as a covariate, but not at the femoral neck. Mendelian models with a residual polygenic component were not excluded. These models were consistent with the presence of a rare Mendelian genotype of prevalence 3–19 %, causing high BMD at the hip and spine in these families, with additional polygenic effects. Total heritability range at the lumbar spine was 61–67 % and at the femoral neck was 44–67 %. Significant differences in correlation of femoral neck and lumbar spine BMD were observed between male and female relative pairs, with male-male comparisons exhibiting stronger lumbar spine BMD correlation than femoral neck, and female-female comparisons having greater femoral neck BMD correlation than lumbar spine. These findings remained true for parent-offspring correlations when menopausal status was taken into account. The recurrence risk ratio for siblings of probands of a Z-score < −2.0 was 5.4 at the lumbar spine and 5.9 at the femoral neck. Conclusions: These findings support gender- and site-specificity of the inheritance of BMD. These results should be considered in the design and interpretation of genetic studies of osteoporosis.

No-reference image quality assessment using modified extreme learning machine classifier

In this paper, we present a machine learning approach to measure the visual quality of JPEG-coded images. The features for predicting the perceived image quality are extracted by considering key human visual sensitivity (HVS) factors such as edge amplitude, edge length, background activity and background luminance. Image quality assessment involves estimating the functional relationship between HVS features and subjective test scores. The quality of the compressed images are obtained without referring to their original images ('No Reference' metric). Here, the problem of quality estimation is transformed to a classification problem and solved using extreme learning machine (ELM) algorithm. In ELM, the input weights and the bias values are randomly chosen and the output weights are analytically calculated. The generalization performance of the ELM algorithm for classification problems with imbalance in the number of samples per quality class depends critically on the input weights and the bias values. Hence, we propose two schemes, namely the k-fold selection scheme (KS-ELM) and the real-coded genetic algorithm (RCGA-ELM) to select the input weights and the bias values such that the generalization performance of the classifier is a maximum. Results indicate that the proposed schemes significantly improve the performance of ELM classifier under imbalance condition for image quality assessment. The experimental results prove that the estimated visual quality of the proposed RCGA-ELM emulates the mean opinion score very well. The experimental results are compared with the existing JPEG no-reference image quality metric and full-reference structural similarity image quality metric.

Heme biosynthesis by the malarial parasite - Import of delta-aminolevulinate dehydrase from the host red cell

The mouse and human malarial parasites, Plasmodium berghei and Plasmodium falciparum, respectively, synthesize heme de novo following the standard pathway observed in animals despite the availability of large amounts of heme, derived from red cell hemoglobin, which is stored as hemozoin pigment, The enzymes, delta-aminolevulinate dehydrase (ALAD), coproporphyrinogen oxidase, and ferrochelatase are present at strikingly high levels in the P, berghei infected mouse red cell in vivo, The isolated parasite has low levels of ALAD and the data clearly indicate it to be of red cell origin. The purified enzyme preparations from the uninfected red cell and the parasite are identical in kinetic properties, subunit molecular weight, cross-reaction with antibodies to the human enzyme, and N-terminal amino acid sequence. Immunogold electron microscopy of the infected culture indicates that the enzyme is present inside the parasite and, therefore, is not a contaminant, The parasite derives functional ALAD from the host and the enzyme binds specifically to isolated parasite membrane in vitro, suggestive of the involvement of a receptor in its translocation into the parasite, While, ALAD, coproporphyrinogen oxidase, and ferrochelatase from the parasite and the uninfected red cell supernatant have identical subunit molecular weights on SDS-polyacrylamide gel electrophoresis and show immunological cross-reaction with antibodies to the human enzymes, as revealed by Western analysis, the first enzyme of the pathway, namely, delta-aminolevulinate synthase (ALAS) in the parasite, unlike that of the red cell host, does not cross-react with antibodies to the human enzyme, However, ALAS enzyme activity in the parasite is higher than that of the infected red cell supernatant. We therefore conclude that the parasite, while making its own ALAS, imports ALAD and perhaps most of the other enzymes of the pathway from the host to synthesize heme de novo, and this would enable it to segregate this heme from the heme derived from red cell hemoglobin degradation, ALAS of the parasite and the receptor(s) involved in the translocation of the host enzymes into the parasite would be unique drug targets.

Synthesis and properties of main chain discotic liquid crystalline polyethers based on rufigallol

Two series of thermotropic main chain discotic liquid crystalline polyethers, PR4m-n, based on rufigallol were prepared starting from the symmetric tetraethers of rufigallol, R4m; m and n represent the number of carbon atoms in the side chain and spacer segment, respectively. The symmetric tetraethers were in turn readily prepared by selective alkylation of rufigallol under controlled phase-transfer conditions. GPC analysis of the polymers suggested that they were all of moderate molecular weights, with M-n varying between 5400 and 17 000. The length of the spacer segment n in these polyethers was systematically varied, and its effect on the phase transition temperatures and the mesophase structure was examined using DSC, polarized light microscopy, and X-ray diffraction. It is noticed that when the spacer lengths are relatively long(n greater than or equal to 2m), the isotropization temperature (TD-i) decreases as the spacer length n increases, an observation that is in accordance with those previously made. However, when the spacer lengths are relatively small (n < 2m), the dependence of TD-i is quite the opposite; TD-i actually increases with an increase in spacer length. Furthermore, X-ray diffraction studies indicate that, in the discotic columnar mesophases that are formed, the columns pack in a hexagonal manner when n greater than or equal to 2m, while they do so in a rectangular lattice when n < 2m, leading to the formation of Dh and Dr mesophases, respectively. Finally, comparison of the discotic polyethers with their low molar mass analogues confirms the role of polymerization in stabilizing the mesophase; while all the polymers exhibit columnar mesophases, some of their low molar mass analogues are not liquid crystalline.

SPSA Based Feature Relevance Estimation For Video Retrieval

With the availability of a huge amount of video data on various sources, efficient video retrieval tools are increasingly in demand. Video being a multi-modal data, the perceptions of ``relevance'' between the user provided query video (in case of Query-By-Example type of video search) and retrieved video clips are subjective in nature. We present an efficient video retrieval method that takes user's feedback on the relevance of retrieved videos and iteratively reformulates the input query feature vectors (QFV) for improved video retrieval. The QFV reformulation is done by a simple, but powerful feature weight optimization method based on Simultaneous Perturbation Stochastic Approximation (SPSA) technique. A video retrieval system with video indexing, searching and relevance feedback (RF) phases is built for demonstrating the performance of the proposed method. The query and database videos are indexed using the conventional video features like color, texture, etc. However, we use the comprehensive and novel methods of feature representations, and a spatio-temporal distance measure to retrieve the top M videos that are similar to the query. In feedback phase, the user activated iterative on the previously retrieved videos is used to reformulate the QFV weights (measure of importance) that reflect the user's preference, automatically. It is our observation that a few iterations of such feedback are generally sufficient for retrieving the desired video clips. The novel application of SPSA based RF for user-oriented feature weights optimization makes the proposed method to be distinct from the existing ones. The experimental results show that the proposed RF based video retrieval exhibit good performance.

An accurate rating aggregation method for generating item reputation

Many websites presently provide the facility for users to rate items quality based on user opinion. These ratings are used later to produce item reputation scores. The majority of websites apply the mean method to aggregate user ratings. This method is very simple and is not considered as an accurate aggregator. Many methods have been proposed to make aggregators produce more accurate reputation scores. In the majority of proposed methods the authors use extra information about the rating providers or about the context (e.g. time) in which the rating was given. However, this information is not available all the time. In such cases these methods produce reputation scores using the mean method or other alternative simple methods. In this paper, we propose a novel reputation model that generates more accurate item reputation scores based on collected ratings only. Our proposed model embeds statistical data, previously disregarded, of a given rating dataset in order to enhance the accuracy of the generated reputation scores. In more detail, we use the Beta distribution to produce weights for ratings and aggregate ratings using the weighted mean method. Experiments show that the proposed model exhibits performance superior to that of current state-of-the-art models.

Bounds on isoperimetric values of trees

Let G = (V, E) be a finite, simple and undirected graph. For S subset of V, let delta(S, G) = {(u, v) is an element of E : u is an element of S and v is an element of V - S} be the edge boundary of S. Given an integer i, 1 <= i <= vertical bar V vertical bar, let the edge isoperimetric value of G at i be defined as b(e)(i, G) = min(S subset of V:vertical bar S vertical bar=i)vertical bar delta(S, G)vertical bar. The edge isoperimetric peak of G is defined as b(e)(G) = max(1 <= j <=vertical bar V vertical bar)b(e)(j, G). Let b(v)(G) denote the vertex isoperimetric peak defined in a corresponding way. The problem of determining a lower bound for the vertex isoperimetric peak in complete t-ary trees was recently considered in [Y. Otachi, K. Yamazaki, A lower bound for the vertex boundary-width of complete k-ary trees, Discrete Mathematics, in press (doi: 10.1016/j.disc.2007.05.014)]. In this paper we provide bounds which improve those in the above cited paper. Our results can be generalized to arbitrary (rooted) trees. The depth d of a tree is the number of nodes on the longest path starting from the root and ending at a leaf. In this paper we show that for a complete binary tree of depth d (denoted as T-d(2)), c(1)d <= b(e) (T-d(2)) <= d and c(2)d <= b(v)(T-d(2)) <= d where c(1), c(2) are constants. For a complete t-ary tree of depth d (denoted as T-d(t)) and d >= c log t where c is a constant, we show that c(1)root td <= b(e)(T-d(t)) <= td and c(2)d/root t <= b(v) (T-d(t)) <= d where c(1), c(2) are constants. At the heart of our proof we have the following theorem which works for an arbitrary rooted tree and not just for a complete t-ary tree. Let T = (V, E, r) be a finite, connected and rooted tree - the root being the vertex r. Define a weight function w : V -> N where the weight w(u) of a vertex u is the number of its successors (including itself) and let the weight index eta(T) be defined as the number of distinct weights in the tree, i.e eta(T) vertical bar{w(u) : u is an element of V}vertical bar. For a positive integer k, let l(k) = vertical bar{i is an element of N : 1 <= i <= vertical bar V vertical bar, b(e)(i, G) <= k}vertical bar. We show that l(k) <= 2(2 eta+k k)

Tropical prawn trawl bycatch of fish and seasnakes reduced by Yarrow Fisheye Bycatch Reduction Device

Reducing unwanted trawl bycatch is actively encouraged in Australia, particularly in prawn trawl fisheries. We tested the performance of a Bycatch Reduction Device, the Yarrow Fisheye, during two periods of commercial fishing operations in Australia's Northern Prawn Fishery, by comparing the catches of paired treatment and control nets. We compared the catch weights of the small fish and invertebrate bycatch, and the commercially important tiger prawns, from 42 trawls in 2002. The Yarrow Fisheye reduced the weight of small bycatch by a mean of 22.7%, with no loss of tiger prawn. We also compared the numbers of seasnakes caught in 41 and 72 trawls during the spring trawling seasons of 2004 and 2005, respectively. The Yarrow Fisheye reduced the catches by a mean of 43.3%. Flume-tank tests of the Yarrow Fisheye showed that this device created a slow water-flow region extending over 2 m downstream from its position in the net, and close to where the catch accumulates. Finfish and seasnakes may be exploiting this slow water-flow region to escape via the eye, Although the reductions in fish and seasnake bycatch were excellent, we think they could be further improved by relating differences in fisheye position and localised water displacements, to design and rigging changes.

Induced smoothing for rank regression with censored survival times

Adaptions of weighted rank regression to the accelerated failure time model for censored survival data have been successful in yielding asymptotically normal estimates and flexible weighting schemes to increase statistical efficiencies. However, for only one simple weighting scheme, Gehan or Wilcoxon weights, are estimating equations guaranteed to be monotone in parameter components, and even in this case are step functions, requiring the equivalent of linear programming for computation. The lack of smoothness makes standard error or covariance matrix estimation even more difficult. An induced smoothing technique overcame these difficulties in various problems involving monotone but pure jump estimating equations, including conventional rank regression. The present paper applies induced smoothing to the Gehan-Wilcoxon weighted rank regression for the accelerated failure time model, for the more difficult case of survival time data subject to censoring, where the inapplicability of permutation arguments necessitates a new method of estimating null variance of estimating functions. Smooth monotone parameter estimation and rapid, reliable standard error or covariance matrix estimation is obtained.

Does the size of subsamples taken from multispecies trawl catches affect estimates of catch composition and abundance?

Although subsampling is a common method for describing the composition of large and diverse trawl catches, the accuracy of these techniques is often unknown. We determined the sampling errors generated from estimating the percentage of the total number of species recorded in catches, as well as the abundance of each species, at each increase in the proportion of the sorted catch. We completely partitioned twenty prawn trawl catches from tropical northern Australia into subsamples of about 10 kg each. All subsamples were then sorted, and species numbers recorded. Catch weights ranged from 71 to 445 kg, and the number of fish species in trawls ranged from 60 to 138, and invertebrate species from 18 to 63. Almost 70% of the species recorded in catches were "rare" in subsamples (less than one individual per 10 kg subsample or less than one in every 389 individuals). A matrix was used to show the increase in the total number of species that were recorded in each catch as the percentage of the sorted catch increased. Simulation modelling showed that sorting small subsamples (about 10% of catch weights) identified about 50% of the total number of species caught in a trawl. Larger subsamples (50% of catch weight on average) identified about 80% of the total species caught in a trawl. The accuracy of estimating the abundance of each species also increased with increasing subsample size. For the "rare" species, sampling error was around 80% after sorting 10% of catch weight and was just less than 50% after 40% of catch weight had been sorted. For the "abundant" species (five or more individuals per 10 kg subsample or five or more in every 389 individuals), sampling error was around 25% after sorting 10% of catch weight, but was reduced to around 10% after 40% of catch weight had been sorted.