Includes abstract. Includes bibliographical references (leaves 98-100).

Convolutional neural networks (CNNs) adapt the regular fully-connected neural network (NN) algorithm to facilitate image classification. Recently, CNNs have been demonstrated to provide superior performance across numerous image classification databases including large natural images (Krizhevsky et al., 2012). Furthermore, CNNs are more readily transferable between different image classification problems when compared to common alternatives. The extension of CNNs to video classification is simple and the rationale behind the components of the model are still applicable due to the similarity between image and video data. Previous CNNs have demonstrated good performance upon video datasets, however have not employed methods that have been recently developed and attributed improvements in image classification networks. The purpose of this research to build a CNN model that includes recently developed elements to present a human action recognition model which is up-to-date with current trends in CNNs and current hardware. Focus is applied to ensemble models and methods such as the Dropout technique, developed by Hinton et al. (2012) to reduce overfitting, and learning rate adaptation techniques. The KTH human action dataset is used to assess the CNN model, which, as a widely used benchmark dataset, facilitates the comparison between previous work performed in the literature. Three CNNs are built and trained to provide insight into design choices as well as allow the construction of an ensemble model. The final ensemble model achieved comparative performance to previous CNNs trained upon the KTH data. While the inclusion of new methods to the CNN model did not result in an improvement on previous models, the competitive result provides an alternative combination of architecture and components to other CNN models.

Includes bibliography. / The thesis deals with graphical display techniques based on the singular value decomposition. These techniques, known as biplots, are used to find low dimensional representations of multidimensional data matrices. The aim of the thesis is to provide a review of biplots for a practical statistician who is not familiar with the area. It therefore focuses on the underlying theory, assuming a standard statisticians' knowledge of matrix algebra, and on the interpretation of the various plots. The topic falls in the realm of descriptive statistics. As such, the methods are chiefly exploratory. They are a means of summarising the data. The data matrix is represented in a reduced number of dimensions, usually two, for simplicity of display. The aim is to summarise the information in the matrix and to present a visual representation of this information. The aim in using graphical display techniques is that the "gain in interpretability far exceeds the loss in information" (Greenacre, 1984). A graphical description is often more easy to understand than a numerical one. Histograms and pie charts are familiar forms of data representation to many people with no other, or very rudimentary, statistical understanding. These are applicable to univariate data. For multivariate data sets, univariate methods do not reveal interesting relationships in the data set as a whole. In addition, a biplot can be presented in a manner which can be readily understood by non-statistically minded individuals. Greenacre (1984) comments that only in recent years has the value of statistical graphics been recognised. Young (1989) notes that recently there has been a shift in emphasis, among statisticians towards exploratory data analysis methods. This school of thought was given momentum by the publication of the book "Exploratory Data Analysis" (Tukey, 1977). The trend has been facilitated by advances in computer technology which have increased both the power and the accessibility of computers. Biplot techniques include the popular correspondence analysis. The original proponents of correspondence analysis (among them Benzecri) reject probabilistic modelling. At the other extreme, some view graphical display techniques as a mere preliminary to the more traditional statistical approaches. Under the latter view, graphical display techniques are used to suggest models and hypotheses. The emphasis in exploratory data techniques such as graphical displays is on 'getting a feel' for the data rather than on building models and testing hypotheses. These methods do not replace model building and hypothesis testing, but supplement them. The essence of the philosophy is that models are suggested by the data, rather than the frequently followed route of first fitting a model. Some work has gone into developing inferential methods, with hypothesis tests and associated p-values for biplot-type techniques (Lebart et al, 1984, Greenacre, 1984). However, this aspect is not important if the techniques are viewed merely as exploratory. Chapter Two provides the mathematical concepts necessary for understanding biplots. Chapter Three explains exactly what a biplot is, and lays the theoretical framework for the biplot techniques that follow. The goal of this chapter is to provide a framework in which biplot techniques can be classified and described. Correlation biplots are described in Chapter Four. Chapter Five discusses the principal component biplot, and the link between these and principal component analysis is drawn. In Chapter Six, correspondence analysis is presented. In Chapter Seven practical issues such as choice of centre are discussed. Practical examples are presented in Chapter Eight. The aim is that these examples illustrate techniques commonly applicable in practice. Evaluation and choice of biplot is discussed in Chapter Nine.

Includes bibliographical references. / A risk budget is the quantification of an acceptable amount of risk that a pension fund is willing to take in its investment decisions, without significantly sacrificing its ability to meet its current and future liabilities.

Includes abstract. Includes bibliographical references (leaves 153-157).

Effective management needs to know sizes of animal populations. This can be accomplished in various ways, but a very popular way is mark-recapture studies. Mark-recapture studies need a way of telling if a captured animal has been previously seen. For traditional mark-recapture, this is achieved by applying a tag to the animal. For non-invasive mark-recapture methods which exploit photographs, there is no tag on the animal’s body. As a result, these methods require animals to be individually identifiable. They assess if an animal has been caught before by examining photographs for animals which have individual-specific marks (Cross et al., 2014; Gomez et al., 2016; Beijbom et al., 2016; Körschens, Barz, and Denzler, 2018). This study develops a model which can reliably match photographs of the same individual based on individual-specific marks. The model consists of two main parts, an object detection model, and a classifier which takes two photos as input and outputs a predicted probability that the pair is from the same individual (a match). The object detection model is a convolutional neural network (CNN) and the matching classifier is a special kind of CNN called a siamese network. The siamese network uses a pair of CNNs that share weights to summarise the images, followed by some dense layers which combine the summaries into measures of similarity which can be used to predict a match. The model is tested on two case studies, humpback whales (HBWs) and western leopard toads (WLTs). The HBW dataset consists of images originally collected by various institutions across the globe and uploaded to the Happywhale platform which encourages scientists to identify individual mammals. HBWs can be identified by their fins and specials markings. There is lots of data for this problem. The WLT dataset consists of images collected by citizen scientists in South Africa. They were either uploaded to iSpot, a citizen science project which collects images or sent to the (WLT) project, a conservation project staffed by volunteers. WLTs can be identified by their unique spots. There is a little data for this problem. One part of this dataset consists of labelled individuals and another part is unlabelled. The model was able to give good results for both HBWs and WLTs. In 95% of the cases the model managed to correctly identify if a pair of images is from the same HBW individual or not. It accurately identified if a pair of images is drawn from the same WLT individual or not in 87% of the cases. This study also assessed the effectiveness of the semi-supervised approach on the WLT unlabelled dataset. In this study, the semisupervised approach has been partially successful. The model was able to identify new individuals and matches which were not identified before, but they were relatively few in numbers. Without an exhaustive check of the data, it is not clear whether this is due to the failure of the semi-supervised approach, or because there are not many matches in the data. After adding the newly identified and labelled individuals to the WLT labelled dataset, the model slightly improved its performance and correctly identified 89% of WLT pairs. A number of computer-aided photo-matching algorithms have been proposed (Matthé et al., 2017). This study also assessed the performance of Wild-ID (Bolger et al., 2012), one of the commonly used photo-matching algorithm on both HBW and WLT datasets. The model developed in this thesis achieved very competitive results compared with Wild-ID. Model accuracies for the proposed siamese network were much higher than those returned by Wild-ID on the HBW dataset, and roughly the same on the WLT dataset.

Includes bibliographical references. / The African penguin (Spheniscus demersus) population in southern Africa has declined from approximately 575 000 adults at the start of the 20th century to 180 000 adults in the early 1990s. The population is still declining, leading to the International Union for the Conservation of Nature upgrading the status of African penguins to Endangered on the Red List of Threatened Species. This dissertation uses a systems dynamics approach to produce a model incorporating all important pressures. The model is stochastic and spatially explicit, and uses expert opinion where data are not available. The model has been produced and revised with the help of the Penguin Modelling Group, based at the University of Cape Town. The modelling process culminated in a workshop where participants experimented with the model themselves. The model in this dissertation is only applicable to the penguin population on Robben Island and, as such, conclusions drawn cannot necessarily be applied to other penguin colonies.

Reinforcement learning is a relatively new and unexplored branch of machine learning with a wide variety of applications. This study investigates reinforcement learning and provides an overview of its application to a variety of different problems. We then explore the possible use of reinforcement learning for telescope target selection and scheduling in astronomy with the hope of effectively mimicking the choices made by professional astronomers. This is relevant as next-generation astronomy surveys will require near realtime decision making in response to high-speed transient discoveries. We experiment with and apply some of the leading approaches in reinforcement learning to simplified models of the target selection problem. We find that the methods used in this study show promise but do not generalise well. Hence while there are indications that reinforcement learning algorithms could work, more sophisticated algorithms and simulations are needed.

Includes bibliographical references (leaves 96-102)

Includes bibliographical references. / The analysis of growth data is important as it allows us to assess how fast things grow and determine various factors that have impact on their growth. In the current study, growth measurements on body features (body mass, wing length, head length, bill (culmen) length, foot length, and tarsus length) for Grey-headed Gulls populating Bonaero Park and Modderfontein Pan in Gauteng province, South Africa, and for Swift Terns on Robben Island were taken. Different methods such as polynomial regressions, non-parametric models and non-linear mixed effects models have been used to fit models to growth data. In recent years, non-linear mixed effects models have become an important tool for growth models. We have fitted univariate inverse exponential, Gompertz, logistic, and Richards non-linear mixed effects models to each of the six body features. We have modeled these six features simultaneously by adding a categorical covariate, which distinguishes between different features, to the model. This approach allows for straightforward comparison of growth between the different body features. In growth studies, the knowledge of the age of each individual is an essential information for growth analysis. For Swift Terns, the exact age of most chicks was unknown, but a small portion of the sample was followed from nestling up to the end of the study period. For chicks with unknown age, we estimated age by fitting the growth curve, obtained from birds with known age, to the mass measurements of the chick with unknown age. It was found that the logistic models were most appropriate to describe the growth of body mass and wing length while the Gompertz models provided best fits for bill, tarsus, head and foot for Grey-headed Gulls. For Swift Terns, the inverse exponential model provided the best univariate fit for four of six features. The logistic model, with a variance function increasing as a power of fitted values, with a different power for each feature and autoregressive correlation structure for within bird errors with errors from different features within the same subject assumed to be independent, gave the best model to describe the growth of all body features taken simultaneously for both Grey-headed Gull and Swift Tern data. It was shown that growth of Grey-headed Gull and Swift Tern chicks occurs in the following order (foot, body mass, tarsus)-(bill, head)-( wing) and (tarsus, foot)-(body mass, bill, head)-(wing) , respectively.