Doctoral Theses
Permanent link for this collection
The Doctoral Theses collection contains digital copies of AUT doctoral theses deposited with the Library since 2004 and made available open access. All theses for doctorates awarded from 2007 onwards are required to be deposited in Tuwhera Open Theses unless subject to an embargo.
For theses submitted prior to 2007, open access was not mandatory, so only those theses for which the author has given consent are available in Tuwhera Open Theses. Where consent for open access has not been provided, the thesis is usually recorded in the AUT Library catalogue where the full text, if available, may be accessed with an AUT password. Other people should request an Interlibrary Loan through their library.
Browse
Recent Submissions
- ItemCan a Spiking Neural Network Predict Unknown Learning Histories Solely on a Snapshot of Binary Choice Data?(Auckland University of Technology, 2023) Plessas, AnnaBehaviour is influenced by past experiences, allowing us to make predictions based on current behavioural patterns. The aim of this doctoral project was to predict (more precisely, to retrodict) unknown learning histories based on small-sized datasets derived from binary choices made by pigeons. These choice behaviour datasets were extracted as discrete events from published research papers and each data included a 5-second window of observable behaviour after the delivery of a reinforcer. A spiking neural network (SNN) with a single Leaky Fire-and-Integrate neuron was developed to process these data and generate retrodictions of learning histories. Experiment 1 showed that retrodictions can be made by approaching behavioural data differently, without manipulating the reinforcer-behaviour relationship. Pigeons' binary choices provided sufficient information to the SNN model, which successfully matched the manual analysis of their actual choice behaviour patterns. The SNN's ability to make retrodictions relied solely on the pigeons' unique response patterns and the SNN's learning capacity. Retrodictions were successful even when the SNN was tested with new samples of various sizes from the same datasets. Thus, the SNN model demonstrated its capability to learn and make accurate predictions from behavioural data. In Experiment 2, the effectiveness of the SNN was validated by comparing its performance with that of other artificial neural networks. Three deep learning models were developed. The retrodictions made by these models were then compared to the performance of the SNN. The results showed that all models were able to accurately retrodict the pigeons' learning history. However, when additional performance measures such as F1 and precision were taken into account, the SNN outperformed all other deep neural network (DNN) models. Experiment 3 showcased the SNN's capability to work with novel small-sized datasets consisting of choice behaviours of other pigeons, both individuals and groups, who had slightly different learning histories. By recalculating the SNN's firing rates in a personalised manner, better predictive performance was achieved compared to conventional approaches used in generalisation tests, despite variations in the pigeons' learning histories. To confirm the method's reliability, Experiment 4 involved retraining the SNN with new datasets by using two transfer-learning techniques (fine-tuning and feature extraction) and then testing it on small new datasets. Both strategies yielded robust retrodictions, demonstrating the advantages of the applied methodologies. This thesis may become a valuable addition to the behaviour-analytic toolbox by providing a tool capable of retrodicting unknown learning histories from limited behavioural datasets containing a small window of binary choices. Collectively, the experiments demonstrate that the SNN is an effective tool for understanding the connection between learning histories and behaviour in behavioural research. The SNN exhibited adaptability and responsiveness to relatively small amounts of data from observable behaviour and produced retrodictions, thus demonstrating its potential to replace labour-intensive manipulations of the reinforcer-behaviour relationship or lengthy common training procedures. It makes personalised predictions possible and facilitates the study of differences in individual learning patterns shedding light on the relation between learning history and behaviour. By retrodicting learning histories, this work establishes a foundation for exploring the use of new training methodologies, using optimal training conditions tailored to individual organisms and specific learning tasks.
- ItemEvaluation of Podiatry Services for Māori in Aotearoa(Auckland University of Technology, 2022) Ihaka, BelindaThis work aimed to determine the effectiveness of Māori diabetes podiatry services in reducing lower limb amputation in Aotearoa by Māori for Māori. The first objective of this study was to evaluate the current evidence regarding the effectiveness of diabetes podiatry services in Aotearoa. The second objective was to explore the views and perceptions through a Māori lens of (i) Māori podiatrists with an Annual Practicing Certificate in Aotearoa who provide diabetes podiatry services (ii) Māori stakeholders who provide services utilising te ao Māori concepts; and (ii) Māori with diabetic foot problems relating to the effectives of podiatric services in Aotearoa. The third objective was to ensure the research benefits Māori and aligns with tikanga values. In reviewing the literature, we considered effectiveness to include reduced length of hospital stay; reduced hospital admission; return to primary care; improved patient self-management; reduction in ulceration/re-ulceration; reduction in amputation; and limb salvage. Only international studies met these criteria. The three studies (Craig et al., 2013; Perrin et al., 2012a; Searle, 2008) clearly demonstrated triaging people based on their foot risk category either in the community or secondary setting is an appropriate way of determining acute from chronic foot pathology. However, there was no clear consensus on how to effectively manage moderate-to-high foot risk categories. A limitation of the literature review was the lack of evidence for diabetes podiatry services in Aotearoa. A kaupapa Māori evaluation approach was used to identify the effectiveness of diabetes podiatry services by Māori for Māori. This approach ensured the process of, and results from the work aligned with tikanga Māori. A collaborative approach between the researcher, the National Hauora Coalition, and a research whānau determined the evaluation processes. A mixed methods approach using semi-structured interviews, electronic surveys and quantitative service data was collected to inform this work. Key themes from the evaluation concluded that current diabetes podiatry services are effective when Māori feel engaged with the practitioner and Māori podiatrists embed mātauranga and tikanga in their approach to Māori with diabetes. The participants in this study suggested mutual learning within culturally safe environments. Furthermore, culturally responsive learning opportunities need to be embedded in undergraduate studies and continuing professional development for registered podiatrists and those wishing to practice in Aotearoa. Finally, funding bodies need to invest in Māori development and capability if they are truly committed to the Pae Ora (Healthy Futures) Bill (2022). It is essential that provisions for the inclusion of non-clinical measures of wellbeing are incorporated into diabetes podiatry practice and reporting. These are necessary to ensure these services are meeting the aspirations of Māori. In conclusion, there is a significant opportunity to embed effective Māori-driven diabetes podiatry service to reduce lower limb amputation among Māori in Aotearoa through authentic and collaborative approaches.
- ItemAnalytic Methods in Finance with Applications to Portfolio and Risk Management(Auckland University of Technology, 2023) Khanthaporn, RewatThe thesis studies and develops an investment portfolio strategy using a regular-vine-based forecasting model in period of the recent COVID-19 crisis. The model parameter estimation technique uses families of Bayesian inference and variational Bayes inference. The optimisation model uses family of machine learning algorithms. Overall, the thesis comprises three papers in which the ultimate outcome in paper three is the solution of dynamic portfolio allocation. Prior to that, the first two papers develop the multivariate asset returns forecasting models using the inference function of margins method and then apply it to the third paper in a portfolio optimisation model. The full details of each paper are provided in their abstracts: Chapter 3 for paper one, Chapter 4 for paper two and Chapter 5 for paper three. A brief outline of the three papers can be stated as follows. The first paper studies a univariate forecasting model using the hybrid of asymmetric generalised autoregressive conditional heteroskedasticity and intertemporal capital asset pricing, with the following innovations: (1) a mixture of two generalised Pareto distributions and a Gaussian distribution; and (2) generalised error distribution. The Griddy Gibbs sampling algorithm in the Bayesian Markov chain Monte Carlo with parallel computing is used for the model parameter estimation. The study demonstrates the proposed model and estimation method through both simulation and empirical experiments among the benchmarks. It proves that the proposed model statistically outperforms competing models in the return forecasting under the conditions of market turmoil during the COVID-19 period. The second paper extends the first paper from the univariate forecasting model to a multivariate forecasting model using high dimensional data and up to 100 dimensions where the comovement model is a regular vine model. The paper initiates a magnitude 13 bivariate copula candidate for the pair structure well-known in the literature of quantitative risk management. While the estimation techniques for the current paper explore another Bayesian Markov chain Monte Carlo, which is random-walk Metropolis-Hasting sampling, and, in Bayesian machine learning, variational Bayes with (and without) latent variables and data augmentation. Both simulation results and empirical results show satisfactory outcomes, since the proposed model and its estimation can outperform the traditional model. The third paper extends multivariate regular vine forecasting model to the problem of dynamic optimal asset allocation in variate optimisation models. The study introduces evolutionary optimisation algorithms, including a genetic algorithm and a clonal selection algorithm, to optimisation problems. There are two main scenarios in optimisation problems which correspond to three model performance indicators: (1) the reward-risk indicator, (2) the diversity indicator, and (3) the convergence indicator. In addition, stock selection analysis is also applied to the optimisation problem. The empirical studies show that the proposed vine-copula-based forecasting model performs well in optimisation problems in terms of performance measures. Furthermore, based on the scenario experiment, the paper mathematically reveals that the financial market dependence structure has been disrupted as if a new normal has been established since the impact of the COVID-19 pandemic.
- ItemUsing Computer Vision to Identify Objects in an Operating Theatre to Support Safer Working(Auckland University of Technology, 2023) Stephen, OkekeThe accurate selection and identification of drugs in the operating rooms (OR) is a priority for every healthcare institution. During drug preparation and administration in the operating theatre, the anesthetists need to correctly select medications to administer them. Giving the wrong medication can lead to very serious consequences. However, anesthetists can make error cases, especially when they are tired or distracted. This project aims to use computer vision to reduce the likelihood of error and increase patient safety during medication administration in operating theatres. Computer vision methods driven by artificial intelligence have outperformed humans in many tedious tasks involving object identification and recognition. Therefore, in this work, a computer vision-based framework is proposed to identify and extract critical information from medication container labels used during anesthesia in the operating theatres, which can be processed to confirm that the correct medication has been selected. The framework is built to automatically generate voice feedback or raise concerns in the event of a potential drug error. The proposed framework will form part of a front-end to the existing anesthetics medication preparation systems and increase the safety of the whole anesthetic workflow. A number of different approaches have been proposed; currently, the project focuses on automatically recognizing the labels on drug ampules and vials using artificial intelligence-powered computer vision methodologies without the need for QR codes or barcodes on the medication ampoules or vials. The framework is tested for accuracy and compatibility with the anesthetic drug preparation, administration workflow, and efficiency. Also, the research investigated ways to ensure the ampule-to-patient flow is safe – for example, by producing compatible syringe embeddable labels and examining other technologies for recording data around the site of injection for the anesthetic record. After the proof-of-concept development, the framework was rigorously tested and validated for usability using real-time procedures. The reliability, accuracy, and processing speed of critical healthcare products, especially those used in a setting such as the operating theatres, are of utmost importance. The proposed framework achieved remarkable accuracy in identifying the anesthetic drug samples with a rapid processing speed below the maximum threshold (1 second) set for the project while exceeding the originally estimated accuracy threshold.
- ItemThe Pressure-Volume Relationship and Hysteresis Loss in Stirling Refrigerators(Auckland University of Technology, 2023) Yang, Ma Renee Danielle FranciscoStirling refrigerators consist of several spaces in which a volume of gas undergoes expansion and compression. This is also known as a ‘gas spring’. In such a space, heat transfer occurs due to the cyclic temperature difference between the working gas and the adjacent walls. This causes cyclic heat dissipation, known as hysteresis loss. Hysteresis loss is one of the many losses within Stirling machines that are not completely understood. This project investigated the underlying mechanisms of hysteresis loss, by examining the relationships between temperature, heat transfer, pressure in single and multiple space Stirling refrigeration systems. This was carried out with a thorough investigation and analysis with mathematical models, a Sage single cylinder model, and single cylinder experimental tests. An experimental validation of a Twinbird 40 W cooler, a beta-type Stirling refrigerator, was also presented. Finally, the Sage model of an alpha Stirling refrigerator and a simple cylinder with regenerator material was used to explore hysteresis loss within the multiple spaces of a Stirling refrigerator. A simple, closed form equation was developed to show the relationship of net P-V work for given pressure and volume amplitudes with a specified pressure phase shift for sinusoidal motion, which worked for both single cylinder and Stirling refrigerator models. The sinusoidal Schmidt equations were used to show that there will always be a pressure phase shift even in ideal situations for any temperature ratio other than 𝜏 = 1. The pressure phase shift is shown to be in both sinusoidal and discrete execution of the Stirling cycle. An effective pressure phase shift in the discrete Stirling cycle is presented, and an applied cycle is discussed to show how the isochoric processes impose this effective phase shift. It was found that there will always be a pressure phase shift if there is a net heat or P-V work transfer, in both multiple and single space systems. The Peclet number is found to be an insufficient quantity to predict hysteresis loss in Stirling refrigerators. Hysteresis loss, or the net heat transfer, is not always from the gas to the wall; in the alpha Stirling model it was found to be a net heat gain from 10 to 20 Hz. It is proposed that hysteresis is not always a loss in multiple space systems which transfer heat or do work. The design implications from the study are to increase the hydraulic diameter where heat transfer is beneficial. Hysteresis loss was found to vary with the working gas at different frequencies. At 10 Hz, the hysteresis loss in a single space experiment with helium as a working gas was five times that of air. However, it is suggested not to base the working gas selection on hysteresis loss minimisation as the regenerator already minimises the effect of hysteresis loss, and the fact that net heat transfer is still required in the heat exchangers. The regenerator and how it reduces heat transfer and therefore hysteresis within Stirling refrigerators was also explored. It was found that the presence of the regenerator ‘isothermalises’ the system by increasing the overall heat transfer amplitude and decreasing the net heat transfer, therefore reducing the pressure phase shift.