Pattern recognition has become an accessible tool in developing advanced adaptive products. The need for such products is not diminishing but on the contrary, requirements for systems that are more and more aware of their environmental circumstances are constantly growing. Feed-forward neural networks are used to learn patterns in their training data without the need to discover by hand the relationships present in the data. However, the problem of estimating the required size of the neural network is still not solved. If we choose a neural network that is too small for a particular given task, the network is unable to "comprehend" the intricacies of the data. On the other hand if we choose a network size that is too big for the given task, we will observe that there are too many parameters to be tuned for the network, or we can fall in the "Curse of dimensionality" or even worse, the training algorithm can easily be trapped in local minima of the error surface. Therefore, we choose to investigate possible ways to find the 'Goldilocks' size for a feed-forward neural network (which is just right in some sense), being given a training set. Furthermore, we used a common paradigm used by the Roman Empire and employed on a wide scale in computer programming, which is the "Divide-et-Impera" approach, to divide a given dataset in multiple sub-datasets, solve the problem for each of the sub-dataset and fuse the results of all the sub-problems to form the result for the initial problem as a whole. To this effect we investigated modular neural networks and their performance.

The very large dimensionality of real world datasets is a challenging problem for classification algorithms, since often many features are redundant or irrelevant for classification. In addition, a very large number of features leads to a high computational time for classification algorithms. Feature selection methods are used to deal with the large dimensionality of data by selecting a relevant feature subset according to an evaluation criterion. The vast majority of research on feature selection involves conventional single-label classification problems, where each instance is assigned a single class label; but there has been growing research on more complex multi-label classification problems, where each instance can be assigned multiple class labels. This thesis proposes three types of new Multi-Label Correlation-based Feature Selection (ML-CFS) methods, namely: (a) methods based on hill-climbing search, (b) methods that exploit biological knowledge (still using hill-climbing search), and (c) methods based on genetic algorithms as the search method. Firstly, we proposed three versions of ML-CFS methods based on hill climbing search. In essence, these ML-CFS versions extend the original CFS method by extending the merit function (which evaluates candidate feature subsets) to the multi-label classification scenario, as well as modifying the merit function in other ways. A conventional search strategy, hill-climbing, was used to explore the space of candidate solutions (candidate feature subsets) for those three versions of ML-CFS. These ML-CFS versions are described in detail in Chapter 4. Secondly, in order to try to improve the performance of ML-CFS in cancer-related microarray gene expression datasets, we proposed three versions of the ML-CFS method that exploit biological knowledge. These ML-CFS versions are also based on hill-climbing search, but the merit function was modified in a way that favours the selection of genes (features) involved in pre-defined cancer-related pathways, as discussed in detail in Chapter 5. Lastly, we proposed two more sophisticated versions of ML-CFS based on Genetic Algorithms (rather than hill-climbing) as the search method. The first version of GA-based ML-CFS is based on a conventional single-objective GA, where there is only one objective to be optimized; while the second version of GA-based ML-CFS performs lexicographic multi-objective optimization, where there are two objectives to be optimized, as discussed in detail in Chapter 6. In this thesis, all proposed ML-CFS methods for multi-label classification problems were evaluated by measuring the predictive accuracies obtained by two well-known multi-label classification algorithms when using the selected featuresม namely: the Multi-Label K-Nearest neighbours (ML-kNN) algorithm and the Multi-Label Back Propagation Multi-Label Learning Neural Network (BPMLL) algorithm. In general, the results obtained by the best version of the proposed ML-CFS methods, namely a GA-based ML-CFS method, were competitive with the results of other multi-label feature selection methods and baseline approaches. More precisely, one of our GA-based methods achieved the second best predictive accuracy out of all methods being compared (both with ML-kNN and BPMLL used as classifiers), but there was no statistically significant difference between that GA-based ML-CFS and the best method in terms of predictive accuracy. In addition, in the experiment with ML-kNN (the most accurate) method selects about twice as many features as our GA-based ML-CFS; whilst in the experiments with BPMLL the most accurate method was a baseline method that does not perform any feature selection, and runs the classifier once (with all original features) for each of the many class labels, which is a very computationally expensive baseline approach. In summary, one of the proposed GA-based ML-CFS methods managed to achieve substantial data reduction, (selecting a smaller subset of relevant features) without a significant decrease in predictive accuracy with respect to the most accurate method.

My original contribution to knowledge is a portfolio of 29 compositions written for the open world videogame Apotheon that constitutes an improvement to the effectiveness and implementation of recombinant videogame music. This portfolio was developed in response to the criticisms of multiple authors such a K. Collins, W. Phillips, S. Huiberts, A. Burnt, D. Raybould, R. Stevens, and others on the negative effects of excessive use of looping in video game music. A collaboration between myself and the programmer and game designer Lee Vermeulen brought the development of a new stochastic recombinant music engine that was implemented in Apotheon. The purpose of the engine is to replace continuous linear looping with unique musical variations in key areas of the game in which players might spend large amounts of time. The music in these areas is dynamically recombined in multiple independent layers using weighted probabilities that are altered according to the development of the narrative. The portfolio was inspired by aleatoric techniques found in 18th century musical dice games, and expanded on stochastic recombinant techniques found in earlier videogame soundtracks such as Ballblazer, Times of Lore and Legend of Zelda: Ocarina of Time.

Additive manufacturing (AM) techniques provide engineering design flexibility not available when manufacturing is constrained by the tool-path restrictions of conventional subtractive techniques such as turning, milling and grinding. AM techniques allow the manufacture of complex form, light weight components with optimised geometries and topographies, including internal and re-entrant features. These features may greatly enhance the components functional capability. The design flexibility may allow a reduction in assembly part count, with a corresponding reduction in assembly time. Additionally, the ability to use high performance engineering metals in the AM process, such as 316 stainless steel, titanium Ti6Al4V and cobalt chrome provide the aerospace, medical and automotive industries with a new manufacturing toolbox using familiar raw materials. These quality-driven industries are fully aware of the potential of AM and are actively engaged and invested with the AM industry and research community. The complex features and design freedom providing great potential for these industries also presents challenges for surface measurement and characterisation. Surface measurement is vital to assure compliance with designed sealing, bearing, flow and adhesion properties of the component. Parts manufactured using AM are not exempt from the stringent quality requirements applicable to other manufacturing processes and so surface texture requirements will be incorporated into drawings and design specifications, imposed by customers onto suppliers. There will need to be a common language and approved standards. Compliance verification will be mandatory. If a feature is specified on a drawing then these industries will require verification that the component complies with design requirements. Traditionally, line-of-sight measuring devices were able to follow the tool pathways to access and measure these surfaces. With the advent of additive processes, new techniques will need to be developed. X-ray computed tomography (CT) has been used successfully for dimensional and defect detection as it allows the measurement of internal and re-entrant features. Thus far, there has been little research on the application of CT for the measurement of surface texture. This thesis reports on the development of a novel technique, detailing the first extraction of areal surface texture parameters per a recognised standard (ISO 25178-2) from CT scans of AM components. Industry will require reproducibility of measurements and so an interlaboratory comparison was performed to compare CT measurement results using this technique from four laboratories. The repeatability and accuracy of surface measurements is also vital for industrial applications and so the influence on extracted surface texture parameter values of selected CT measurement and reconstruction factors has been investigated. Extraction of true 3D data from CT requires the generation of new surface characterisation parameters to take full advantage of the technique and a new parameter has been developed to enable the true surface of re-entrant surfaces to be characterised. The additive process itself is complex and verification of consistent additive machine performance is vital for production. A series of small, inexpensive, surfacespecific measurement artefacts has been developed and built to characterise the build chamber and provide production process verification. This series of inter-related experimental investigations were chosen to be industrially relevant, to be linked closely to component function and be used as practical measurement and surface characterisation techniques. This work is intended, as far as possible, to not be machine-specific, but to be applicable to all CT machines and all metal powder bed fusion (PBF) AM machines. As AM and CT machine capability improves, as it inevitably will, the techniques and applications presented here are designed to evolve with these changes.

This thesis addresses YouTube-based programming in Saudi Arabia. It focuses on the bottom-up, professionally generated content (PGC) produced by the Saudi youth exclusively for YouTube between 2010 and 2016. It explores the economic, political and cultural context of YouTube-based programming and is aimed at understanding how the programme makers operate within such a restrictive political and media environment, and the interaction between the content creators, the audiences and the Saudi government. The thesis examines the perspective of the content creators by utilizing ten semi-structured online interviews, and the view of the Saudi audience by employing seven semi-structured offline focus group interviews with twenty-nine participants in Glasgow, Scotland. The interviews were conducted and transcribed in Arabic, then translated into English. This analysis leads to the emergence of four primary themes: limited political and cultural empowerment for young Saudis offline; strong financial empowerment for the content creators; the government as the most powerful actor in the Saudi media scene; and the relevance of the content to the Saudi audience as the main driver for the popularity of YouTube-based programming. The analysis provides insights and implications for policy and practices, arguing that the concept and practices of cultural policy are humble in Saudi Arabia and that there is an urgent need for a political decision to establish a nationwide cultural policy to promote and govern the arts and culture. This policy needs to incorporate both online and offline creativity, including YouTube-based programming.

The research reported in this thesis is aimed at developing and demonstrating the performance of a p-i-n / vertical cavity enhanced (VCE) photodetector structure with different material compositions of dilute nitride. The tunable selectivity is taken into account during the development in order for the VCE photodetector to be able to perform as a tunable receiver. In the first part, the performance of p-i-n photodetectors with 10 and 20 quantum wells are presented. The 20 quantum well photodetector exhibits the highest ever reported quantum efficiency of 80% at -5V bias for applications at 1.0µm wavelength. However, variation in the absorption’s thickness equates to a trade-off between speed, bandwidth, noise-equivalent-power, and detectivity. A GaInNAs/GaAs VCE photodetector with an internal gain of 1.55 for room temperature operation at 1.3µm wavelength is next demonstrated. This is the first internal gain ever reported using a GaInNAs VCE photodetector with a 5nm FWHM and an overall quantum efficiency of 67%. The proposed VCE photodetector of GaInNAs/GaNAs showed impressive multiplication at a low reverse bias of 0.5V. The internal gain is detected to be 2.45 at -2V bias with a responsivity of 1.63A/W for room temperature operation. The responsivity and gain of the sample can be enhanced by increasing small range of temperature but then suffered from temperature-induced broadening of the full width half maximum (FWHM). Finally, an avalanche VCE photodetector is proposed, where the preliminary outcome shows the current oscillation behaviour under illumination at room temperature operation. The sample produces very low responsivity of 2.15mA/W due to growth problems with the charge layer which affect the electric field in the multiplication region.

Atlas-Based Segmentation of medical images is an image analysis task which involves labelling a desired anatomy or set of anatomy from images generated by medical imaging modalities. The overall goal of atlas-based segmentation is to assist radiologists in the detection and diagnosis of diseases. By extracting the relevant anatomy from medical images and presenting it in an appropriate view, their work-flow can be optimised. This portfolio-style thesis discusses the research projects carried out in order to evaluate the applicability of atlas-based methods to a variety of medical imaging problems. The thesis describes how atlas-based methods have been applied to heart segmentation, to extract the heart for further cardiac analysis from cardiac CT images, to kidney segmentation, to prepare the kidney for automated perfusion measurements, and to coronary vessel tracking, in order to improve on the quality of tracking algorithms. This thesis demonstrates how state of the art atlas-based segmentation techniques can be applied successfully to a range of clinical problems in different imaging modalities. Each application has been tested using not only standard experimentation principles, but also by clinically-trained personnel to evaluate its efficacy. The success of these methods is such that some of the described applications have since been deployed in commercial products. While exploring these applications, several techniques based on published literature were explored and tailored to suit each individual application. This thesis describes in detail the methods used for each application in turn, recognising the state of the art, and outlines the author's contribution in every application.

This thesis is a study of the affect of differences in the density of a fluid on the mixing regions of jets, which may be laminar or turbulent. These differences in density are present for three main reasons, namely; when the speed of the fluid is of the same order of magnitude as the local speed of sound; when there are large temperature differences in the fluid; and when the fluid consists of a mixture of components the relative proportions of which vary from point to point. Three problems are considered. These are: the flow far from the orifice of a plane and of a round jet and the mixing region on the surface of the core of a plane jet near the orifice. This last problem is idealised as the mixing of two semi-infinite streams. For flows of jet type, the assumption of a coefficient of eddy kinematic viscosity in turbulent flow leads to the possibility of combining in one the equations for laminar and turbulent motion. The method used is to expand the stream function in a Rayleigh-Jansen series. The first term of this series corresponds to the stream function when the fluid is of constant density. The series is developed in powers of a small parameter whose magnitude depends on the density differences in the fluid. Only the second term of this series is found explicitly. This term gives the first order effect that changes in density have on the flow. The solutions of all examples considered are, with on exception, given in analytical form. The last appendix to the the thesis shows the connection between Stewartson's (1957) approach to the problem of finding uniformly valid approximate solutions to the boundary layer equations and Lighthill's (1948) method. This connection is shown by working out one of the problems considered by Stewartson, namely, the wake past a flat plate, using Lighthill's method.

Laser ablation has been investigated as a possible technique for the contactless deflection of Near Earth Asteroids. It is achieved by irradiating the surface of an asteroid with a laser light source. The absorbed heat from the laser beam sublimates the surface, transforming the illuminated material directly from a solid to a gas. The ablated material then forms into a plume of ejecta. This acts against the asteroid, providing a controllable low thrust, which pushes the asteroid away from an Earth-threatening trajectory. The potential of laser ablation is dependent on understanding the physical and chemical properties of the ablation process. The ablation model is based on the energy balance of sublimation and was developed from three fundamental assumptions. Experimental verification was used to assess the viability of the ablation model and its performance in inducing a deflection action. It was achieved by ablating a magnesium-iron silicate rock, under vacuum, with a 90 W continuous wave laser. The laser operated at a wavelength of 808 nm and provided intensities that were below the threshold of plasma formation. The experiment measured the average mass flow rate, divergence geometry and temperature of the ejecta plume and the contaminating effects - absorptivity, height and density - of the deposited ejecta. Results were used to improve the ablation model. A critical discrepancy was in the variation between the previously predicted and experimentally measured mass flow rate of the ablated ejecta. Other improvements have also included the energy absorption within the Knudsen layer, the variation of sublimation temperature with local pressure, the temperature of the target material and the partial re-condensation of the ablated material. These improvements have enabled the performance of the ablation process and the specifications of the laser to be revised. Performance exceeded other forms of electric propulsion that provided an alternative contactless, low thrust deflection method. The experimental results also demonstrated the opportunistic potential of laser ablation. Using existing technologies, with a high technology readiness level, a small and low-cost mission design could demonstrate the technologies, approaches and synergies of a laser ablation mission. The performance of the spacecraft was evaluated by its ability to deflect a small and irregular 4 m diameter asteroid by at least 1 m/s. It was found to be an achievable and measurable objective. The laser ablation system could be successfully sized and integrated into a conventional solar-power spacecraft. Mission mass and complexity is saved by the direct ablation of the asteroid's surface. It also avoids any complex landing and surface operations. Analysis therefore supports the general diversity and durability of using space-based lasers and the applicability of the model's experimental verification.

Nanotopographical patterning of biomaterial substrates has great potential for biofunctionalisation of devices for clinical applications, such as in orthopaedics. Nanotopography comprising 120 nm diameter nanopits with a partially disordered arrangement of up to +/- 50 nm offset from a square lattice with 300 nm centre to centre spacing (NSQ50, fabricated by electron beam lithography) has been characterized as being osteogenic. Following the finding of osteogenesis of mesenchymal stromal cells derived from human bone marrow (MSCs) on the NSQ50 nanotopography, MSCs cultured on ε-polycaprolactone (PCL) embossed with the NSQ50 pattern was used for this study on molecular mechanisms underlying NSQ50 induced MSC osteogenesis: the functional coupling of gene expression and osteogenesis, the molecular regulatory events driving gene expression and osteogenesis, and the possible link of metabolomics with molecular signalling of MSCs on the NSQ50 surface. Temporal analysis of gene expression for MSCs on the NSQ50 surface revealed that MSC fate commitment and osteogenic differentiation was transcriptionally controlled. The cell cycle and growth regulating transcription factor C-MYC was found to be significantly repressed, whereas the osteogenic transcription factor RUNX2 was up-regulated at 5 days of cell culture, and this was followed upregulation of the osteoblast specific transcription factor osterix (OSX) at days 11 and 13. Following this transcription factor activation, osteoblast specific marker genes were induced with increased alkaline phosphatase (ALP) observed at day 16, increased osteopontin (OPN) at day 20 and increased osteocalcin (OCN) at day 28. These data suggested that transcription factors regulated MSC osteogenic commitment at the early stage, and induced osteogenic specific marker gene expression at the late stages of cell culture on the NSQ50 surface, resulting in osteogenesis of the MSCs. Signalling pathway analysis illustrated that bone morphogenetic protein 2 (BMP2) was the initial signalling molecule that triggered osteogenic differentiation of MSCs by inducing RUNX2 expression via the canonical SMAD pathway. BMP2 and its transmembrane receptor type 1A (BMPR1A) were stimulated by 3 nanotopographical cues by 3 days of cell culture on the NSQ50 surface, whereas the induction of other transmembrane receptors, including the low density lipoprotein-receptor related protein 5 (LRP5) and integrin subunits α3, α4, β1, and β3 were not observed. Inhibition of BMP2 signalling by the BMP2 antagonist noggin resulted in down-regulation of RUNX2 and ALP. Further analysis of BMP2 signalling revealed that BMP2 also modulated expression of the microRNA (miR)- 23b which targets RUNX2. The effect of BMP2 signalling on the expression of RUNX2 was enhanced by co-localizing with integrin αvβ5 (the vitronectin (VN) receptor) which was found to be up-regulated after 5 days cell culture. Metabolomics data for MSCs on the NSQ50 surface during early osteogenic differentiation was analysed. MSC cellular metabolite analysis revealed possible changes in bioenergetic balance with shifts towards more mitochondrial oxidative process, possibly indicating a switch in MSCs on the surface towards lineage-specific commitment. Further analysis of the metabolomics data illustrated PPARG ligands from the polyunsaturated fatty acid family was downregulated, suggesting the inhibition of adipocyte differentiation in MSCs on the surface. The down-regulation of unsaturated fatty acids could also be involved in the regulation of Ca2+ channels which positively regulate BMP2 expression. The metabolomics data, together with gene expression and signalling pathway analysis demonstrated that MSCs on the NSQ50 surface initiated osteogenic commitment after 3 days of cell culture, with BMP2 initiating osteogenic transcription factor stimulation of mature and functional osteoblasts on the surface.