Search

This thesis is a study of data selection and placement in heterogeneous memories in modern high-performance computer architectures. Memory systems are becoming increasingly complex and diverse, which complicates the search for optimal data placement and reduces the portability of applications. As we enter the dawn of the exascale era, memory models have to be rethought to consider the new trade-offs between latency, bandwidth, capacity, persistence and accessibility, and their impact on performance. Moreover, this data management needs to be simplified and brought within reach of domain scientists in fields outside of Computer Science. To address this issue, this work focuses on studying data movement, data optimisation and memory management in systems with heterogeneous memory. Firstly, a new algorithm was developed that improves the computation of data exchange in the context of multigrid data redistribution. Secondly, multiple APIs for memory management were unified into a single abstraction that provides memory allocations and transfers in the form of a portable, adaptive and modular library. Lastly, the allocation management was studied in a high-level language along with ways to enable low-level control over memory placement for a high-level language. The Adjacent Shifting of PEriodic Node data (ASPEN) algorithm, presented in this thesis, provides better performance than state-of-the-art algorithms used for producer-consumer data redistribution of block-cyclic organised data, as used in distributed numerical applications and libraries (e.g. ScaLAPACK). The MAMBA library was developed and aims to facilitate data management on heterogeneous memory systems. It uses a data broker developed with library cooperation and interoperability in mind. In addition to providing portability and memory abstraction, it also serves as a comparison tool for benchmarking or exploratory experiments. Finally, a use case of memory management in C for a Python application based on a distributed framework has been studied as a proof-of-concept for providing direct memory management to high-level application development. This work presents a data-centric approach to the challenges heterogeneous memory creates for performance-seeking applications.

Supercomputers continue to expand both in size and complexity as we reach the beginning of the exascale era. Networks have evolved, from simple mechanisms which transport data to subsystems of computers which fulfil a significant fraction of the workload that computers are tasked with. Inevitably with this change, assumptions which were made at the beginning of the last major shift in computing are becoming outdated. We introduce a new latency-bandwidth model which captures the characteristics of sending multiple small messages in quick succession on modern networks. Contrary to other models representing the same effects, the pipelining latency-bandwidth model is simple and physically based. In addition, we develop a discrete-event simulation, Fennel, to capture non-analytical effects of communication within models. AllReduce operations with small messages are common throughout supercomputing, particularly for iterative methods. The performance of network operations are crucial to the overall time-to-solution of an application as a whole. The Message Passing Interface standard was introduced to abstract complex communications from application level development. The underlying algorithms used for the implementation to achieve the specified behaviour, such as the recursive doubling algorithm for AllReduce, have to evolve with the computers on which they are used. We introduce the recursive multiplying algorithm as a generalisation of recursive doubling. By utilising the pipelining nature of modern networks, we lower the latency of AllReduce operations and enable greater choice of schedule. A heuristic is used to quickly generate a near-optimal schedule, by using the pipelining latency-bandwidth model. Alongside recursive multiplying, the endpoints of collective operations must be able to handle larger numbers of incoming messages. Typically this is done by duplicating receive queues for remote peers, but this requires a linear amount of memory space for the size of the application. We introduce a single-consumer multipleproducer queue which is designed to be used with MPI as a protocol to insert messages remotely, with minimal contention for shared receive queues.

With the development of new technologies and their increased applications in the context of a local government, cities have started to claim that they are smart. Smart Cities make use of Information and Communication Technologies (ICTs) to support planning and policy making. For an appropriate and sustainable functioning of these smart cities, collecting data about the different aspects of their territory and operations, including its citizens, is a crucial activity. Currently, there are two main avenues in which smart cities can collect data about their citizens: either through sensors, and cameras strategically placed throughout the city or by asking citizens to voluntarily donate to the local government their personal data (i.e., citizen engagement or 'e-participation'). Despite the growth and increasing prevalence of the latter practice, little attention has been given to how individuals experience the risks of data donation. Often, studies consider data donation as an aspect of the phenomenon of surveillance, or as a type of data sharing. This study theorises and empirically examines data donation and its risks as a phenomenon which is separate from either surveillance or data sharing. Focusing on mobility data, this study draws on two established donation and privacy risk frameworks to investigate how the risks of donating personal data to a smart city are experienced and socially constructed. The thematic analysis of ten focus groups conducted showed that, in the context of this empirical examination, privacy-specific risks alone do not constitute constructed risks. Instead, they combine in various ways with perceived donation risks to constitute more nuanced and embedded risk constructions. Donation risks are seen as potential consequences of privacy risks and combined they constitute the risks of donating data. This thesis underlines the importance of the context under which data donation takes place as well as privacy's value in a free and democratic society.

This thesis explores the issue of capitalist exploitation of digital media where free time and creativity are fundamental elements in the production of digital goods. The thesis focuses in particular on free open source hardware communities in which hackers give up a considerable amount of free or leisure time and creativity to produce open technology. Hardware hacking represents relatively new model designing and assembling of hardware based on commons-based peer production (CBPP). In this research, I examine how free time and creativity can be exploited in open source communities, with corporations benefiting from community wealth. I investigate how free or leisure time becomes a regime of "hyper-exploitation" from which capital is increasingly accumulated. It is hyper-exploitation precisely because, whereas workers receive wages in return to their labour, producers in CBPP are unpaid (Ritzer, 2014). I focus on two Free Open Source Hardware (FOSHW) projects: RepRap and Arduino, as example cases in this thesis. Discussions on RepRap and Arduino mailing lists and the data from interviews are analysed. I explore how the free time and creativity of volunteers are exploited in FOSHW communities and how hackers react to capitalist exploitation. This thesis shows that hackers have differing aims and motivations in RepRap and Arduino communities. The discussions on the issue of open source, the issue of self-replicating, the issue of customisation, fun in the RepRap community and the issue of Arduino clones all provide a basis for analysing the logic of hyper-exploitation. Thjs in turn is based on the commercialisation of open hardware goods and the exploitation of voluntary labour which plays a key role in the production and distribution of software and design. As a basis for further discussion, I introduce a set of key concepts that play an important role in the analysis of capitalist exploitation in Free Open Source Hardware (FOSHW) communities. These concepts include free time, creativity, capitalist exploitation, democracy, hacker, open source communities, collaboration, work, free labour and fun. I also discuss how the line between free time and work time is blurred Page 4   and the production of open source software and design has come to be seen as part of a free time activity. This thesis shows that free time and activity of volunteers in open source communities are exploited by technology companies and the term "fun" may disguise capitalist exploitation, in which the line between leisure time and work time is not clear. Creative activities taking place in free time create value that is appropriated by companies (see Kostakis and Bauwens, 2014). Even though hackers have fun when developing software and design, the efforts and creativity of hackers can be viewed as productive labour and therefore turn into capital in the market. The collaborative relationship between the firms and open source communities may enable capitalists to make a profit from peer production. I will discuss this issue in more detail in the thesis. The story of MakerBot in Chapter 5 and the issue of Arduino trademark in Chapter 6 provide us with important information enabling us to discuss in depth the hyper-exploitation of voluntary labour. I will scrutinise the concept of hyper-exploitation in Chapter 2. Open source communities, on the one hand, allow humans to participate in the production of open technology. This can be also understood as the democratisation of production. Voluntary labour, on the other hand, may be appropriated by firms. In this thesis, I explore this contradiction in FOSHW communities. In this research, I undertake interviews and collect data from the RepRap and Arduino mailing lists. I apply corpus text analysis and thematic analysis to the mailing lists and interview data. Two different empirical cases are used in the research. Firstly, Replicating Rapid Prototyper (RepRap) has been chosen in relation to the production and manufacturing of free-open-source 3D printers. RepRap is known as a self-replicating machine that produces most of its own components. Secondly, Arduino, itself a low-power open-source single-board computer has been selected as an empirical case.

There is an increasing number of ICT systems (e.g. to communicate, do business, vote, control industrial processes or critical infrastructures, etc.) whose security depends intrinsically on human users. Concomitantly, there are many reported critical vulnerabilities that are due to users failing to follow security procedures or to behave as ICT scientists have decided is appropriate. A solution to this problem will only be found by addressing it radically differently, by treating it as a true socio-technical problem rather than just a technical one. We must understand how the technical components (e.g., software processes and digital communication protocols) and the social components (e.g., user interaction processes and user behaviour) of a system interoperate, and thus consider the system as a true socio-technical system, with people at its heart. This requires extending the technical analysis approaches with a mature understanding of human behaviour, as humans are complicated and nothing guarantees that, even if they learned how to operate a technology, either from a manual or through its use, they will comply with what they learned. Reasons include cognitive biases, fallacies, ignorance, distraction, laziness, curiosity of different uses, insufficient awareness of the security sensitivity of their behaviour, etc. This thesis focuses on developing an innovative methodology to analyse the sociotechnical security of ICT systems. To advance the state-of-the-art to the point where the wide spectrum of socio-technical security features of systems can be modelled formally and automatically analysed, this thesis aims to: (i) design a methodology to tackle the socio-technical security of systems; (ii) define a formal modelling language expressive enough to cover the diverse security features of socio-technical systems; (iii) define libraries of prototypical socio-technical security properties, behavioural user models, socio-technical attack/threat models; (iv) implement a toolkit, an integrated front-end to holistically conduct formal security analysis of socio-technical systems; (v) demonstrate a proof-of-concept on a number of archetypal case studies.

Analysing failures to improve the reliability of high performance computing systems and data centres is important. The primary source of information for diagnosing system failures is the system logs and it is widely known that finding the cause of a system failure using only system logs is incomplete. Resource utilisation data – recently made available – is another potential useful source of information for failure analysis. However, large High-Performance Computing (HPC) systems generate a lot of data. Processing the huge amount of data presents a significant challenge for online failure diagnosis. Most of the work on failure diagnosis have studied errors that lead to system failures only, but there is little work that study errors which lead to a system failure or recovery on real data. In this thesis, we design, implement and evaluate two failure diagnostics frameworks. We name the frameworks CORRMEXT and EXERMEST. We implement the Data Type Extraction, Feature Extraction, Correlation and Time-bin Extraction modules. CORRMEXT integrates the Data Type Extraction, Correlation and Time-bin Extraction modules. It identifies error cases that occur frequently and reports the success and failure of error recovery protocols. EXERMEST integrates the Feature Extraction and Correlation modules. It extracts significant errors and resource use counters and identifies error cases that are rare. We apply the diagnostics frameworks on the resource use data and system logs on three HPC systems operated by the Texas Advanced Computing Center (TACC). Our results show that: (i) multiple correlation methods are required for identifying more dates of groups of correlated resource use counters and groups of correlated errors, (ii) the earliest hour of change in system behaviour can only be identified by using the correlated resource use counters and correlated errors, (iii) multiple feature extraction methods are required for identifying the rare error cases, and (iv) time-bins of multiple granularities are necessary for identifying the rare error cases. CORRMEXT and EXERMEST are available on the public domain for supporting system administrators in failure diagnosis.

Automata learning is a popular technique used to automatically construct an automaton model from queries, and much research has gone into devising specific adaptations of such algorithms for different types of automata. This thesis presents a unifying approach to many existing algorithms using category theory, which eases correctness proofs and guides the design of new automata learning algorithms. We provide a categorical automata learning framework---CALF---that at its core includes an abstract version of the popular L* algorithm. Using this abstract algorithm we derive several concrete ones. We instantiate the framework to a large class of Set functors, by which we recover for the first time a tree automata learning algorithm from an abstract framework, which moreover is the first to cover also algebras of quotiented polynomial functors. We further develop a general algorithm to learn weighted automata over a semiring. On the one hand, we identify a class of semirings, principal ideal domains, for which this algorithm terminates and for which no learning algorithm previously existed; on the other hand, we show that it does not terminate over the natural numbers. Finally, we develop an algorithm to learn automata with side-effects determined by a monad and provide several optimisations, as well as an implementation with experimental evaluation. This allows us to improve existing algorithms and opens the door to learning a wide range of automata.

In view of the importance of effective comparison of uncertain data in real-world applications, this thesis focuses on developing new similarity measures with high accuracy. As it identifies and articulates an inherent limitation of the popular set-theoretic similarity measures for continuous intervals where they return the same similarity value for very different sets of intervals (termed as aliasing), this thesis first underpins a new axiomatic definition of a robust similarity measure and then proposes a new similarity measure for continuous intervals based on their bidirectional subsethood. Beyond establishing theoretical foundation of the new measure, the thesis also demonstrates its robust results vis-a-vis existing measures and suitability for real world applications. In the next stage, it develops a generalized framework to assess similarity between discontinuous intervals as current approaches involve loss of discontinuity information and are also affected by aliasing of the popular measures— these weaknesses impact the accuracy of similarity results. This thesis further integrates Allen’s theory with the new generalized framework to make the latter more efficient. Moving beyond intervals, this thesis extends the new similarity measure both vertically and horizontally (α-cut based) for comparing type-1 (T1) fuzzy sets as the shortcoming of popular similarity measures persists with their extension to T1 fuzzy sets. The empirical evaluation of the extended new measures with respect to key existing fuzzy set-theoretic similarity measures shows that the vertically extended new measure behaves intuitively for various types of fuzzy sets, except for non-normal fuzzy sets; however, the α-cut based extended new measure meets expectation in all cases. At the final stage, the utility of the new similarity measure is explored to improve the robustness of fuzzy integral (FI) based uncertain-data aggregation. As existing approaches to generate fuzzy measures (FMs) rely on popular similarity measures to capture the degree of similarity among individual sources (and their combinations), they are also impacted by their aforesaid limitation. Therefore, this thesis develops a new FM based on the new similarity measure which can generate intuitive aggregation outcome when used in combination with an FI.

We compare the category theoretic semantics for binding signatures by Power and Tanaka with the abstract approach to universal algebra by Hyland. It is striking to see that two different ideas turn out to be so similar. We especially note that both approaches rely heavily on considering a monoid in the monoidal structure induced by a 0-cell in the Kleisli bicategory generated by a pseudo-distributive law of pseudo-monads. We further explain the implications the discovery of those similarities have by considering constructions that were only used in either of the two bodies of work.

Last-Level Cache (LLC) represents the bulk of a modern CPU processor's transistor budget and is essential for application performance as LLC enables fast access to data in contrast to much slower main memory. Problematically, technology constraints make it infeasible to scale LLC capacity to meet the ever-increasing working set size of the applications. Thus, future processors will rely on effective cache management mechanisms and policies to get more performance out of the scarce LLC capacity. Applications with large working set size often exhibit streaming and/or thrashing access patterns at LLC. As a result, a large fraction of the LLC capacity is occupied by dead blocks that will not be referenced again, leading to inefficient utilization of the LLC capacity. To improve cache efficiency, the state-of-the-art cache management techniques employ prediction mechanisms that learn from the past access patterns with an aim to accurately identify as many dead blocks as possible. Once identified, dead blocks are evicted from LLC to make space for potentially high reuse cache blocks. In this thesis, we identify variability in the reuse behavior of cache blocks as the key limiting factor in maximizing cache efficiency for state-of-the-art predictive techniques. Variability in reuse prediction is inevitable due to numerous factors that are outside the control of LLC. The sources of variability include control-flow variation, speculative execution and contention from cores sharing the cache, among others. Variability in reuse prediction challenges existing techniques in reliably identifying the end of a block's useful lifetime, thus causing lower prediction accuracy, coverage, or both. To address this challenge, this thesis aims to design robust cache management mechanisms and policies for LLC in the face of variability in reuse prediction to minimize cache misses, while keeping the cost and complexity of the hardware implementation low. To that end, we propose two cache management techniques, one domain-agnostic and one domain-specialized, to improve cache efficiency by addressing variability in reuse prediction. In the first part of the thesis, we consider domain-agnostic cache management, a conventional approach to cache management, in which the LLC is managed fully in hardware, and thus the cache management is transparent to the software. In this context, we propose Leeway, a novel domain-agnostic cache management technique. Leeway introduces a new metric, Live Distance, that captures the largest interval of temporal reuse for a cache block, providing a conservative estimate of a cache block's useful lifetime. Leeway implements a robust prediction mechanism that identifies dead blocks based on their past Live Distance values. Leeway monitors the change in Live Distance values at runtime and dynamically adapts its reuse-aware policies to maximize cache efficiency in the face of variability. In the second part of the thesis, we identify applications, for which existing domain-agnostic cache management techniques struggle in exploiting the high reuse due to variability arising from certain fundamental application characteristics. Specifically, applications from the domain of graph analytics inherently exhibit high reuse when processing natural graphs. However, the reuse pattern is highly irregular and dependent on graph topology; a small fraction of vertices, hot vertices, exhibit high reuse whereas a large fraction of vertices exhibit low- or no-reuse. Moreover, the hot vertices are sparsely distributed in the memory space. Data-dependent irregular access patterns, combined with the sparse distribution of hot vertices, make it difficult for existing domain-agnostic predictive techniques in reliably identifying, and, in turn, retaining hot vertices in cache, causing severe underutilization of the LLC capacity. In this thesis, we observe that the software is aware of the application reuse characteristics, which, if passed on to the hardware efficiently, can help hardware in reliably identifying the most useful working set even amidst irregular access patterns. To that end, we propose a holistic approach of software-hardware co-design to effectively manage LLC for the domain of graph analytics. Our software component implements a novel lightweight software technique, called Degree-Based Grouping (DBG), that applies a coarse-grain graph reordering to segregate hot vertices in a contiguous memory region to improve spatial locality. Meanwhile, our hardware component implements a novel domain-specialized cache management technique, called Graph Specialized Cache Management (GRASP). GRASP augments existing cache policies to maximize reuse of hot vertices by protecting them against cache thrashing, while maintaining sufficient flexibility to capture the reuse of other vertices as needed. To reliably identify hot vertices amidst irregular access patterns, GRASP leverages the DBG-enabled contiguity of hot vertices. Our domain-specialized cache management not only outperforms the state-of-the-art domain-agnostic predictive techniques, but also eliminates the need for any storage-intensive prediction mechanisms.

Validated integration is a family of methods that compute enclosures for sets of initial conditions in the Initial Value Problems. The Taylor model based validated integration methods use truncated Taylor series to approximate the solution to the Initial Value Problem and often give better results than other validated integration methods. Validated integration methods, and especially Taylor model based ones, become increasingly more impractical as the number of variables in the system get higher. In this thesis, we develop techniques that mitigate the issues related to the dimension of the system in Taylor model based validated integration methods. This is done by taking advantage of the compositional structure of the problem when possible. More precisely, the main contribution of this thesis is to enable computing an enclosure to a higher dimensional system by using enclosures for smaller lower dimensional subsystem that are contained in the larger system. The techniques called shrink wrapping and preconditioning are used in the Taylor model based validated integration to improve accuracy. We also analyse these techniques from a compositional viewpoint and present their compositional counterparts. We accompany compositional version of the Taylor model based validated integration with implementation of our tool CFlow* and experiments using our tool. The experimental results show performance gains for some systems with non-trivial compositional structure. This work was motivated by interest in formally analysing biological systems and we use biological systems examples in a number of our systems.

Inexpensive mobile devices offer new capabilities for non-specialist use in the field for the purpose of conservation. This thesis explores the potential for such devices to be used by citizen scientists interacting with bioacoustic data such as birdsong. This thesis describes design research and field evaluation, in collaboration with conservationists and educators, and technological artefacts implemented as mobile applications for interactive educational gaming and creative composition. This thesis considers, from a participant-centric collaborative design approach, conservationists' demand for interactive artefacts to motivate engagement in citizen science through gameful and playful interactions. Drawing on theories of motivation, frequently applied to the study of Human-Computer Interaction (HCI), and on approaches to designing for motivational engagement, this thesis introduces a novel pair of frameworks for the analysis of technological artefacts and for assessing participant engagement with bioacoustic citizen science from both game interaction design and citizen science project participation perspectives. This thesis reviews current theories of playful and gameful interaction developed for collaborative learning, data analysis, and ground-truth development, describes a process for design and analysis of motivational mobile games and toys, and explores the affordances of various game elements and mechanics for engaging participation in bioacoustic citizen science. This thesis proposes research into progressions for scaffolding engagement with citizen science projects where participants interact with data collection and analysis artefacts. The research process includes the development of multiple designs, analyses of which explore the efficacy of game interactions to motivate engagement through interaction progressions, given proposed analysis frameworks. This thesis presents analysed results of experiments examining the usability of, and data-quality from, several prototypes and software artefacts, in both laboratory conditions and the field. This thesis culminates with an assessment of the efficacy of proposed design analysis frameworks, an analysis of designed artefacts, and a discussion of how these designs increase intrinsic and extrinsic motivation for participant engagement and affect resultant bioacoustic citizen science data quantity and quality.

The scaling limitations of multi-core processor development have led to a diversification of the processor cores used within individual computers. Heterogeneous computing has become widespread, involving the cooperation of several structurally different processor cores. Central processor (CPU) cores are most frequently complemented with graphics processors (GPUs), which despite their name are suitable for many highly parallel computations besides computer graphics. Furthermore, deep learning accelerators are rapidly gaining relevance. Many applications could profit from heterogeneous computing but are held back by the surrounding software ecosystems. Heterogeneous systems are a challenge for compilers in particular, which usually target only the increasingly marginalised homogeneous CPU cores. Therefore, heterogeneous acceleration is primarily accessible via libraries and domain-specific languages (DSLs), requiring application rewrites and resulting in vendor lock-in. This thesis presents a compiler method for automatically targeting heterogeneous hardware from existing sequential C/C++ source code. A new constraint programming method enables the declarative specification and automatic detection of computational idioms within compiler intermediate representation code. Examples of computational idioms are stencils, reductions, and linear algebra. Computational idioms denote algorithmic structures that commonly occur in performance-critical loops. Consequently, well-designed accelerator DSLs and libraries support computational idioms with their programming models and function interfaces. The detection of computational idioms in their middle end enables compilers to incorporate DSL and library backends for code generation. These backends leverage domain knowledge for the efficient utilisation of heterogeneous hardware. The constraint programming methodology is first derived on an abstract model and then implemented as an extension to LLVM. Two constraint programming languages are designed to target this implementation: the Compiler Analysis Description Language (CAnDL), and the extended Idiom Detection Language (IDL). These languages are evaluated on a range of different compiler problems, culminating in a complete heterogeneous acceleration pipeline integrated with the Clang C/C++ compiler. This pipeline was evaluated on the established benchmark collections NPB and Parboil. The approach was applicable to 10 of the benchmark programs, resulting in significant speedups from 1.26× on “histo” to 275× on “sgemm” when starting from sequential baseline versions. In summary, this thesis shows that the automatic recognition of computational idioms during compilation enables the heterogeneous acceleration of sequential C/C++ programs. Moreover, the declarative specification of computational idioms is derived in novel declarative programming languages, and it is demonstrated that constraint programming on Single Static Assignment intermediate code is a suitable method for their automatic detection.

Finding globally-optimal solutions in quadratic nonconvex optimisation problems deterministically at scale requires exploiting problem structure. Moreover, engineering/industrial applications often present combinatorial aspects, with a nonconvexity interplay from both nonlinearity and integrality constraints. The structure this interplay introduces is difficult to exploit due to different existing mathematical/algorithmic toolboxes for nonlinear-continuous versus discrete-polyhedral optimisation. This thesis addresses two arising challenges: specific commercially-relevant pooling problems that bundle bilinear nonconvexities with a topological and polyhedral structure; semidefinite relaxations (of nonlinear nonconvexity) that integrate with difficulty into polyhedral-based Branch & Cut global solvers. First, we parametrically study pooling structure and explicitly identify sparsity via dominant active topologies under relaxed flow availability for single quality instances. We associate sparse active topological patterns with explicit piecewise objective functions, validating a long-held and heuristically successful intuition. We formally derive strongly-polynomial solutions for several single quality pooling problem subclasses, including some previously-studied nonconvex instances. The conditions in which sparse strongly-polynomial piecewise structure vanishes due to combinatorial complexity has further implications for pooling relaxations in global solvers. Second, we develop an effective lightweight linear outer-approximation of semidefinite relaxations, which we show can easily integrate into global solvers. Compared to previous work, our proposed cuts are sparser in the number of row nonzeros and explicitly selected to improve the objective. We explore relevant engineering trade-offs for sparsity patterns on quadratic programming with box constraints, showing they may immediately extend to quadratically constrained instances. A neural network estimator is key to selecting which strong cuts to generate using objective structure: ranking each cut by expected objective improvement involves solving many semidefinite optimisation problems, an expensive proposition at each Branch & Cut node. The estimator, trained a priori of any instance to solve, predicts objective improvements, taking the computation offline as an application of machine learning in cut selection and global optimisation.

With the rapid development of mobile network and smartphones, the audio-visual communication has achieved great success. The degree of immersion, however, is limited as physical interaction is not allowed. Introduction of vibrotactile data can significantly improve immersive experience of telecommunication and augment conventional multimedia. While acquisition, compression, display and quality evaluation of audio-visual data have been well developed over the past decades, the research on vibrotactile data has just started recently. The main objective of this thesis is to design quality assessment, data compression algorithms and applications for vibrotactile data. As quality assessment is required for evaluating the performance of a codec, we develop the vibrotactile quality assessment prior to designing the vibrotactile codec. This thesis first presents a subjective evaluation protocol dedicated to vibrotactile data. Based on this protocol, subjective experiments are conducted to evaluate the performance of two common objective metrics, i.e. signal-to-noise ratio (SNR) and structural similarity (SSIM), in the vibrotactile domain. The results demonstrate that both of them have high correlation with human vibrotactile perception, so that SNR and SSIM can be used to evaluate the quality of vibrotactile data. Considering a realistic scenario where distortion varies with time, we propose a hybrid objective metric that composites SNR and SSIM. Subjective tests show that the hybrid metric outperforms SNR and SSIM. Next, we propose a vibrotactile data compression algorithm in the spirit of Weber's law. As humans are more sensitive to intensity change when the amplitude of vibrotactile data is low, we apply a companding function to the vibrotactile data. In this way, the quantisation step of high amplitude is larger than that of low amplitude. The curve of the companding function is optimised through a datadriven approach. Subjective experiments are conducted to assess the performance of the proposed vibrotactile codec in terms of human perceived quality. The results demonstrate that no degradation is perceived with a compression ratio of 75%. Furthermore, the latency is much shorter and the computational complexity is much lower than the state-of-the-art. Moreover, two vibrotactile coding methods to represent letters, numbers, symbols, etc. for the visually impaired are proposed. We utilise a vibrotactile display to deliver consecutive vibrational signals to the users and manipulate the frequency and duration of the vibrational signals to code different information. To assess the performance of the vibrotactile coding method, subjective tests are conducted. The results demonstrate that participants are able to learn the codes in about 6-8 hours and recognise words and symbols at an accuracy over 90%. More importantly, the cost of the vibrotactile display is significantly less than that of the commonly-used braille display.

This thesis presents a novel parallel boundary conforming method for simulating two-dimensional incompressible flows on adaptive moving unstructured meshes. The proposed method is especially helpful for studying immiscible multiphase flow problems as the interfaces are represented explicitly by mesh lines, and the grids are able to conform the movements of the interfaces. The fluid flow equations are studied based on the Arbitrary Lagrangian-Eulerian formulation and the Finite Element Method (Li, 2013); parallel linear solvers provided by PETSc are used to solve the discretised incompressible Navier-Stokes equations. There are several important advantages of the boundary conforming method on unstructured meshes. The method is accurate in computing curvature and in resolving the dynamic boundary condition at the interface. It is also able to handle large interface deformation. Due to the complexities of the mesh generator/optimiser, the method only exists in sequential version in the literature, and its application range is severely limited by the computing time and memory of a single processor. The motivation of this work is to take the advantage of the parallel computing platform to extend the sequential method to a parallel method. This will enormously enlarge the application range of the method so that more realistic multiphase flow problems can be investigated. The two major difficulties for the parallel adaptive moving mesh method are: (1) the consistent local remeshing at the inter-processor boundaries between processors; and (2) the dynamic load balancing due to mesh adaptation. To address these issues, a parallel fast marching method is developed such that the characteristic mesh length can be computed globally and this length also determines the condition for local remeshing. The synchronised local remeshing algorithm on inter-process boundaries is then developed in order to preserve consistent partition boundaries. Repartition of the adapted mesh is achieved by ParMETIS and further mesh migration is performed to enforce contiguous sub-meshes. The parallel Laplacian smoothing method is performed to relocate each interior vertex to the arithmetic mean of its neighbouring incident vertices. The parallel adaptive moving mesh is validated by moving bodies with prescribed velocity imposed on the dynamic interfaces. The accuracy of our method increases quadratically as the increase in the initial number of vertices on the interfaces. The discretisations of the incompressible Navier-Stokes equations for multiphase flows are studied; the parallel vectors and matrices resulted from FEM are formed via PETSc linear algebra implementations. Arange of computational fluid dynamics problems involving flow past a circular cylinder, an oscillating drop/bubble and a rising bubble are used to examine the accuracy of the parallel flow solver on both fixed and moving meshes. The parallel performance of the adaptive moving mesh method, flow solver and flow solver on adaptive moving meshes is investigated according to strong and weak scaling analyses. It is found that the adaptive moving mesh method does not scale well but it only takes little computation effort (approximately 2% of the total running time) when studying multiphase flow problems. Most of the computation effort is spent on assembly and solving the linear system. A favourable speedup is observed for multiphase flow problems and a super-linear speedup is achieved up to 32 processors. Generally, for a fixed size problem the best performance of the parallel multiphase flow solver occurs when there are approximately 13000 elements per processor. The code is finally used to simulate 64 rising bubbles. This thesis concludes with a discussion of further improvements and future work to be undertaken.

Extracting valuable information from the visualisation of biological data and turning it into a network model is the main challenge addressed in this thesis. Biological networks are mathematical models that describe biological entities as nodes and their relationships as edges. Because they describe patterns of relationships, networks can show how a biological system works as a whole. However, network inference is a challenging optimisation problem impossible to resolve computationally in polynomial time. Therefore, the computational biologists (i.e. modellers) combine clustering and heuristic search algorithms with their tacit knowledge to infer networks. Visualisation can play an important role in supporting them in their network inference workflow. The main research question is: "How can visualisation support modellers in their workflow to infer networks from biological data?" To answer this question, it was required to collaborate with computational biologists to understand the challenges in their workflow and form research questions. Following the nested model methodology helped to characterise the domain problem, abstract data and tasks, design effective visualisation components and implement efficient algorithms. Those steps correspond to the four levels of the nested model for collaborating with domain experts to design effective visualisations. We found that visualisation can support modellers in three steps of their workflow. (a) To select variables, (b) to infer a consensus network and (c) to incorporate information about its dynamics. To select variables (a), modellers first apply a hierarchical clustering algorithm which produces a dendrogram (i.e. a tree structure). Then they select a similarity threshold (height) to cut the tree so that branches correspond to clusters. However, applying a single-height similarity threshold is not effective for clustering heterogeneous multidimensional data because clusters may exist at different heights. The research question is: Q1 "How to provide visual support for the effective hierarchical clustering of many multidimensional variables?" To answer this question, MLCut, a novel visualisation tool was developed to enable the application of multiple similarity thresholds. Users can interact with a representation of the dendrogram, which is coordinated with a view of the original multidimensional data, select branches of the tree at different heights and explore different clustering scenarios. Using MLCut in two case studies has shown that this method provides transparency in the clustering process and enables the effective allocation of variables into clusters. Selected variables and clusters constitute nodes in the inferred network. In the second step (b), modellers apply heuristic search algorithms which sample a solution space consisting of all possible networks. The result of each execution of the algorithm is a collection of high-scoring Bayesian networks. The task is to guide the heuristic search and help construct a consensus network. However, this is challenging because many network results contain different scores produced by different executions of the algorithm. The research question is: Q2 "How to support the visual analysis of heuristic search results, to infer representative models for biological systems?" BayesPiles, a novel interactive visual analytics tool, was developed and evaluated in three case studies to support modellers explore, combine and compare results, to understand the structure of the solution space and to construct a consensus network. As part of the third step (c), when the biological data contain measurements over time, heuristics can also infer information about the dynamics of the interactions encoded as different types of edges in the inferred networks. However, representing such multivariate networks is a challenging visualisation problem. The research question is: Q3 "How to effectively represent information related to the dynamics of biological systems, encoded in the edges of inferred networks?" To help modellers explore their results and to answer Q3, a human-centred crowdsourcing experiment took place to evaluate the effectiveness of four visual encodings for multiple edge types in matrices. The design of the tested encodings combines three visual variables: position, orientation, and colour. The study showed that orientation outperforms position and that colour is helpful in most tasks. The results informed an extension to the design of BayePiles, which modellers evaluated exploring dynamic Bayesian networks. The feedback of most participants confirmed the results of the crowdsourcing experiment. This thesis focuses on the investigation, design, and application of visualisation approaches for gaining insights from biological data to infer network models. It shows how visualisation can help modellers in their workflow to select variables, to construct representative network models and to explore their different types of interactions, contributing in gaining a better understanding of how biological processes within living organisms work.

Communication may be seen as a process of sending and accepting information among individuals. It is a vital part of emergency response management, sharing the information of situations, victims, family and friends, rescue organisations and others. The obtained contextual information during a disaster event, however, is often dynamic, partial and may be conflicting with each other. Current communication strategies and solutions for emergency response have limitations - in that they are often designed to support information sharing between organisations and not individuals. As a result, they are often not personalisable. They also cannot make use of opportunistic resources, e.g. people nearby the disaster-struck areas that are ready to help but are not a part of any organisation. However, history has told us such people are often the first responders that provide the most immediate and useful help to the victims. On the other hand, the advanced and rich capabilities of mobile smartphones have become one of the most interesting topics in the field of mobile technologies and applied science. It is especially interesting when it can be expanded to become an effective emergency response tool to discover affected people and connect them with the first responders and their families, friends and communities. At present, research on emergency response is ineffective for handling large-scale disasters where professional rescuers could not reach victims in disaster struck-areas immediately. This is because current approaches are often built to support formal emergency response teams and organizations. Individual emergency response efforts, e.g. searching for missing people (inc. families and friends), are often web-based applications that are also not effective. Other works focus on sensory development that lacks integrated search and rescue approaches. In this thesis, I developed a distributed and personalisable Mobile Kit Disaster Assistant (MKA) system that is underpinned by a formal foundation. It aims at gathering emergency response information held by multiple resources before, during and after a large-scale disaster. As a result, contextual and background information based on a formal framework would be readily available, if a disaster indeed strikes. To this end, my core contribution is to provide a structural formal framework to encapsulate important information that is used to support emergency response at a personal level. Several (conceptual) structures were built to allow an individual to express his/her own individual circumstances, inc. relationships with others and health status that will determine how he/she may communicate with others. The communication framework is consisting of several new components: a rich and holistic Emergency Response Communication Framework, a newly developed Communication and Tracking Ontology (CTO), a newly devised Emergency Response Agent Communication Language (ER-ACL) and a brand-new Emergency Response Agent Communication Protocol (ER-ACP). I have framed the emergency response problem as a multi-agent problem where each smartphone would act as an agent for its user; each user would take on a role depending on requirements and/or the tasks at hand and the above framework is aimed to be used within a peer to peer distributed multiagent system (MAS) to assist emergency response efforts. Based on this formal framework, I have developed a mobile application, the MKA system, to capture important features of EM and to demonstrate the practicalities and value of the proposed formal framework. This system was carefully evaluated by both domain experts and potential users of targeted user groups using both qualitative and quantitative approaches. The overall results are very encouraging. Evaluators appreciated the importance of the tool and believe such tools are vital in saving lives – that is applicable for large-scale disasters as well as for individual life-critical events.

To obtain good predictions in the presence of imbalance classes has posed significant challenges in the data science community. Imbalanced classed data is a term used to describe a situation where there are unequal number of classes or groups in datasets. In most real-life datasets one of the classes are always higher in number than others and is called the majority class, while the smaller classes are called the minority class. During classifications even with very high accuracy, the classified minority groups are usually very small when compared to the total number of minority in the datasets and more often than not, the minority classes are what is being sought. This work is specifically concern with providing techniques to improve classifications performance by eliminating or reducing negative effects of class imbalance. Real-life datasets have been found to contain different types of error in combination with class imbalance. While these errors are easily corrected, but the solutions to class imbalance have remained elusive. Previously, machine learning (ML) technique has been used to solve the problems of class imbalanced. There are notable shortcomings that have been identified while using this technique. Mostly, it involve fine-tuning and changing parameters of the algorithms and this process is not standardised because of countless numbers of algorithms and parameters. In general, the results obtained from these unstandardised (ML) technique are very inconsistent and cannot be replicated with similar datasets and algorithms. We present a novel technique for dealing with imbalanced classes called variance ranking features selection, that enables machine learning algorithms to classify more of minority classes during classification, hence reducing the negative effects of class imbalance. Our approaches utilised the intrinsic property of the datasets called the variance. As the variance is one of the measures of central tendency of the data items concentration within the datasets vector space. We demonstrated the selections of features at different level of performance threshold thereby providing an opportunity for performance and feature significance to be assessed and correlated at different levels of prediction. In the evaluations we compared our features selections with some of the best known features selections techniques using proximity distance comparison techniques and verify all the results with different datasets, both binary and multi classed with varying degree of class imbalance. In all the experiments, the results we obtained showed a significant improvement when compared with other previous work in class imbalance.

Hardware has become more prone to faults, due to wear-out and faults caused during the manufacturing process. The reliability of hardware is becoming more dependent on the ability to continually adapt to faults and current fault tolerant approaches are susceptible to faults. A computational model of biological self-repair in the brain, derived from observing the distributed role of astrocytes (a glial cell found in the mammalian brain), has captured self-repair within neural networks; these are known as neuro-glia networks. Astrocytes have been shown to facilitate biological self-repair in silent or near silent neurons in the brain by increasing the Probability of Release (PR) in healthy synapses. Astrocytes modulate synaptic activity, which leads to increased or decreased PR. To date, this has been proven with computational modelling and therefore, the next step is to replicate this self-repair process in hardware to provide self-repairing electronic information processing systems. A key challenge for hardware neuro-glia networks implementation is the facilitation of scalable communication between interacting neurons and astrocyte cells. There are large volumes of neurons/astrocytes with different communication patterns and this network is viewed as a two-tiered network: 1. High speed temporal spike event (neural network); 2. Low speed numerical inositol trisphosphate information exchange (astrocyte network). This thesis addresses the key challenges of providing scalable communication for a neuro-glia network with low-level Networks-on-Chip (NoC) topologies. This network supports astrocyte to neuron/synapse communication at a local level and astrocyte communication at a global level i.e. the astrocyte network. The astrocyte process is inherently slow, thus a ring topology exploits this slow change and sacrifices high throughput for a low area overhead, this is analogous to the astrocyte process. This astrocyte was applied in hardware and results demonstrate that novel ring topology provides a trade-off between low area/interconnect wiring overhead whilst supporting realistic communication speeds for both the slow-changing data between astrocytes and the higher throughput neuron networks.

Fuzzy inference system provides an effective means for representing and processing vagueness and imprecision. Conventional fuzzy modelling requires either complete experts' knowledge or given datasets to generate rule bases such that the input spaces can be fully covered. Although fuzzy interpolation enhances the power of conventional fuzzy inference approaches by addressing the problem of lack of knowledge represented in the rule bases, it is still difficult for real-world applications to obtain sufficient experts' knowledge and/or data to generate a sparse rule base to support fuzzy interpolation. Also, the generated rule bases are usually fixed and therefore cannot support dynamic situations. In addition, all existing fuzzy interpolation approaches were developed based on the Mamdani fuzzy model, which are not applicable for the TSK fuzzy model. It significantly restricts the applicability of the TSK fuzzy inference systems. This PhD work, in the first part, presents a novel fuzzy inference approach, termed "TSK+ fuzzy inference", to address the issue of performing the TSK inference over sparse rule bases. The proposed TSK+ fuzzy inference approach extends the conventional TSK fuzzy inference by considering the degree of similarity between given inputs and corresponding rule antecedents instead of conventional overlapped match degree, which allows TSK inference to be performed over sparse rule bases, dense rule bases, and imbalanced rule bases. In order to support the proposed TSK+ inference approach, a data-driven rule base generation method is also presented in this work. In addition, the proposed TSK+ inference approach has been further extended to deal with interval type-2 fuzzy sets. The effectiveness of this system in enhancing the TSK fuzzy inference is demonstrated through two real-world applications: a network intrusion detection system, and a network quality of service management system. In addition, in the second part of this work, a new rule base generation and adaptation method is developed in order to relax the requirement of rule base generation, which allows the fuzzy rule base to be generated with minimal or even without a priori knowledge. The proposed method mimics the pedagogic approach of experiential learning, which achieves automatic rule base generation and adaptation by transferring the proceeding performance experiences when performing inferences. The proposed rule base generation and adaptation method has been evaluated by not only a mathematical model but also a well-known control problem, inverted pendulum. The experimental results show that the control system can generate an applicable rule base to support the system running, thus demonstrating the effectiveness of the proposed approach.

In today's world, because of developments in medical sciences, people are living longer, particularly in the advanced countries. This increasing of the lifespan has caused the prevalence of age-related diseases like Alzheimer's and dementia. Researches show that ion channel disruptions, especially the formation of permeable pores to cations by Aβ plaques, play an important role in the occurrence of these types of diseases. Therefore, early detection of such diseases, particularly using non-invasive tools can aid both patients and those scientists searching for a cure. To achieve the goal toward early detection, the computational analysis of ion channels, ion imbalances in the presence of Aβ pores in neurons and fault detection is done. Any disruption in the membrane of the neuron, like the formation of permeable pores to cations by Aβ plaques, causes ionic imbalance and, as a result, faults occur in the signalling of the neuron. The first part of this research concentrates on ion channels, ion imbalances and their impacts on the signalling behaviour of the neuron. This includes investigating the role of Aβ channels in the development of neurodegenerative diseases. Results revealed that these types of diseases can lead to ionic imbalances in the neuron. Ion imbalances can change the behaviour of neuronal signalling. Therefore, by identifying the pattern of these changes, the disease can be detected in the very early stages. Then the role of coupling and synchronisation effects in such diseases were studied. After that, a novel method to define minimum requirements for synchronicity between two coupled neurons is proposed. Further, a new computational model of Aβ channels is proposed and developed which mimics the behaviour of a neuron in the course of Alzheimer's disease. Finally, both fault computation and disease detection are carried out using a residual generation method, where the residuals from two observers are compared to assess their performance.

Complex systems are a rapidly increasing area of research covering numerous disciplines including economics and even cancer research, as such the optimisation of the simulations of these systems is important. This thesis will look specifically at two cellular automata based growth models the Eden growth model and the Invasion Percolation model. These models tend to be simulated storing the cluster within a simple array. This work demonstrates that for models which are highly sparse this method has drawbacks in both the memory consumed and the overall runtime of the system. It demonstrates that more modern data structures such as the HSH tree can offer considerable benefits to these models. Next, instead of optimising the software simulation of the Eden growth model, we detail a memristive-based cellular automata architecture that is capable of simulating the Eden growth model called the MEden model. It is demonstrated that not only is this method faster, up to 12; 704 times faster than the software simulation, it also allows for the same system to be used for the simulation of both EdenB and EdenC clusters without the need to be reconfigured; this is achieved through the use of two different parameters present in the model Pmax and Pchance. Giving the model a broader range of possible clusters which can aid with Monte-Carlo simulations of the model. Finally, two methods were developed to be able to identify a difference between fractally identical clusters; connected component labelling and convolution neural networks are the methods used to achieve this. It is demonstrated that both of these methods allow for the identification of individual Eden clusters able to classify them as either an EdenA, EdenB, or EdenC cluster, a highly nontrivial matter with current methods. It is also able to tell when a cluster was not an Eden cluster even though it fell in the fractal range of an Eden cluster. These features mean that the verification of a new method for the simulation of the Eden model could now be automated.

Traditional robotics have the capabilities of the robot hard coded and have the robot function in structured environments (structured environments are those that are predefined for a given task). This approach can limit the functionality of a robot and how they can interact in an environment. Behaviour networks are reactive systems that are able to function in unstructured dynamic environments by selecting behaviours to execute based on the current state of the environment. Behaviour networks are made up of nodes that represent behaviours and these store an activation value to represent the motivation for that behaviour. The nodes receive inputs from a variety of sources and pass proportions of that input to other nodes in the network. Behaviour networks traditionally also have their capabilities predefined. The main aim of this thesis is to expand upon the concepts of traditional robotics by demonstrating the use of distributed behaviours in an environment. This thesis aims to show that distributing object specific data, such as; behaviours and goals, will assist in the task planning for a mobile robot. This thesis explores and tests the traditional behaviour network with a variety of experiments. Each experiment showcases particular features of the behaviour network including flaws that have been identified. Proposed solutions to the found flaws are then presented and explored. The behaviour network is then tested in a simulated environment with distributed behaviours and the dynamic behaviour network is defined. The thesis demonstrates that distributed behaviours can expand the capabilities of a mobile robot using a dynamic behaviour network.

We analyse transformations between cryptographic primitives and, for each transformation, we do two studies: its provable security (proving that if the original cryptographic primitive is secure, then the transformed cryptographic primitive is also secure); its proof presentation (exploring improved ways of presenting the proof). Our contributions divide into two sets: security proofs (sometimes new proofs and sometimes variants of known proofs); proof presentations (inspired by our security proofs) and extraction of lessons learned from them.

Securing Digital Imaging and Communications in Medicine (DICOM) images are essential to protect the privacy of patients, especially in the era of communications, telemedicine and eHealth/mHealth. This increases the demand for rapid security. Nevertheless, a limited number of research work has been conducted to ensure the security of DICOM images while minimising the processing time. Hence, this thesis introduces a novel selective encryption approach to reduce the processing time and sustain robustness of security. The proposed approach selects regions within medical images automatically in the spatial domain using pixel thresholding segmentation technique, then compresses and encrypts them using different encryption algorithms based on their importance. The presented approaches in this thesis are developed for multi-frame and single-frame DICOM images. As the medical images of the same modality and the same body part for different patients have the same statistical distribution, it is possible to segment images from the same modality and same body part into Region of Interest (ROI) and Region of Background (ROB) using a pre-defined threshold. In the case of small ROI, an adaptive two-region encryption approach is applied to single and multi-frame DICOM image. In which, the ROB is encrypted using a light encryption algorithm, while the ROI is encrypted using a sophisticated encryption algorithm. For multi-frame DICOM images, additional time saving has been achieved using one segmentation map based on a pre-defined reference frame for all the DIOCM frames. On the other hand, if ROI is very large, i.e. Mammography DICOM images, the ROI is further split into three regions (multi-region) based on potential security threats, using a novel mathematical model that guarantees shorter encryption time in comparison with the Naive encryption approach. After that, each region is compressed using lossless compression approach, and then encrypted using a different encryption algorithm with different key lengths based on its pixels' intensity. Results from the presented approaches show that processing time is reduced by an average of 54% in comparison with the Naïve encryption approach while maintaining the security of the DICOM images. Further, cryptanalysis metrics are utilised to evaluate the proposed approaches, which indicate good robustness against wide varieties of attacks.

In software engineering, formal methods are mathematical-based techniques that are used in the specification, development and verification of algorithms and programs in order to provide reliability and robustness of systems. One of the most difficult challenges for software engineering is to tackle the complexity of algorithms and software found in concurrent systems. Networked systems have come to prominence in many aspects of modern life, and therefore software engineering techniques for treating concurrency in such systems has acquired a particular importance. Algorithms in the software of concurrent systems are used to accomplish certain tasks which need to comply with the properties required of the system as a whole. These properties can be broadly subdivided into `safety properties', where the requirement is `nothing bad will happen', and `liveness properties', where the requirement is that `something good will happen'. As such, specifying network algorithms and their safety and liveness properties through formal methods is the aim of the research presented in this thesis. Since temporal logic has proved to be a successful technique in formal methods, which have various practical applications due to the availability of powerful model-checking tools such as the NuSMV model checker, we will investigate the specification and verification of network algorithms using temporal logic and model checking. In the first part of the thesis, we specify and verify safety properties for network algorithms. We will use temporal logic to prove the safety property of data consistency or serializability for a model of the execution of an unbounded number of concurrent transactions over time, which could represent software schedulers for an unknown number of transactions being present in a network. In the second part of the thesis, we will specify and verify the liveness properties of networked flooding algorithms. Considering the above in more detail, the first part of this thesis specifies a model of the execution of an unbounded number of concurrent transactions over time in propositional Linear Temporal Logic (LTL) in order to prove serializability. This is made possible by assuming that data items are ordered and that the transactions accessing these data items respects this order, as then there is a bound on the number of transactions that need to be considered to prove serializability. In particular, we make use of recent work which places such bounds on the number of transactions needed when data items are accessed in order, but do not have to be accessed contiguously, i.e., there may be `gaps' in the data items being accessed by individual transactions. Our aim is to specify the concurrent modification of data held on routers in a network as a transactional model. The correctness of the routing protocol and ensuring safety and reliability then corresponds to the serializability of the transactions. We specify an example of routing in a network and the corresponding serializability condition in LTL. This is then coded up in the NuSMV model checker and proofs are performed. The novelty of this part is that no previous research has used a method for detecting serializablity and cycles for unlimited number of transactions accessing the data on routers where the transactions way of accessing the data items on the routers have a gap. In addition to this, linear temporal logic has not been used in this scenario to prove correctness of the network system. This part is very helpful in network administrative protocols where it is critical to maintain correctness of the system. This safety property can be maintained using the presented work where detection of cycles in transactions accessing the data items can be detected by only checking a limited number of cycles rather than checking all possible cycles that can be caused by the network transactions. The second part of the thesis offers two contributions. Firstly, we specify the basic synchronous network flooding algorithm, for any fixed size of network, in LTL. The specification can be customized to any single network topology or class of topologies. A specification for the termination problem is formulated and used to compare different topologies with regards to earlier termination. We give a worked example of one topology resulting in earlier termination than another, for which we perform a formal verification using the NuSMV model checker. The novelty of the second part comes in using linear temporal logic and the NuSMV model checker to specify and verify the liveness property of the flooding algorithm. The presented work shows a very difficult scenario where the network nodes are memoryless. This makes detecting the termination of network flooding very complicated especially with networks of complex topologies. In the literature, researchers focussed on using testing and simulations to detect flooding termination. In this work, we used a robust technique and a rigorous method to specify and verify the synchronous flooding algorithm and its termination. We also showed that we can use linear temporal logic and the model checker NuSMV to compare synchronous flooding termination between topologies. Adding to the novelty of the second contribution, in addition to the synchronous form of the network flooding algorithm, we further provide a formal model of bounded asynchronous network flooding by extending the synchronous flooding model to allow a sent message, non-deterministically, to either be received instantaneously, or enter a transit phase prior to being received. A generalization of `rounds' from synchronous flooding to the asynchronous case is used as a unit of time to provide a measure of time to termination, as the number of rounds taken, for a run of an asynchronous system. The model is encoded into temporal logic and a proof obligation is given for comparing the termination times of asynchronous and synchronous systems. Worked examples are formally verified using the NuSMV model checker. This work offers a constraint-based methodology for the verification of liveness properties of software algorithms distributed across the nodes in a network.

Learning and assessing the Structured Query Language (SQL) is an important step in developing students' database skills. However, due to the increasing numbers of students learning SQL, assessing and providing detailed feedback to students' work can be time consuming and prone to errors. The main purpose of this research is to reduce or remove as many of the repetitive tasks in any phase of the assessment process of SQL statements as possible to achieve the consistency of marking and feedback on SQL answers. This research examines existing SQL assessment tools and their limitations by testing them on SQL questions, where the results reveal that students must attaint essential skills to be able to formulate basic SQL queries. This is because formulating SQL statements requires practice and effort by students. In addition, the standard steps adopted in many SQL assessment tools were found to be insufficient in successfully assessing our sample of exam scripts. The analysis of the outcomes identified several ways of solving the same query and the categories of errors based on the common student mistakes in SQL statements. Based on this, this research proposes a semi-automated assessment approach as a solution to improve students' SQL formulation process, ensure the consistency of SQL grading and the feedback generated during the marking process. The semi-automatic marking method utilities both the Case-Based Reasoning (CBR) system and Rule-Based Reasoning (RBR) system methodologies. The approach aims to reduce the workload of marking tasks by reducing or removing as many of the repetitive tasks in any phase of the marking process of SQL statements as possible. It also targets the improvement of feedback dimensions that can be given to students. In addition, the research implemented a prototype of the SQL assessment framework which supports the process of the semi-automated assessment approach. The prototype aims to enhance the SQL formulation process for students and minimise the required human effort for assessing and evaluating SQL statements. Furthermore, it aims to provide timely, individual and detailed feedback to the students. The new prototype tool allows students to formulate SQL statements using the point-and-click approach by using the SQL Formulation Editor (SQL-FE). It also aims to minimise the required human effort for assessing and evaluating SQL statements through the use of the SQL Marking Editor (SQL-ME). To ensure the effectiveness of the SQL-FE tool, the research conducted two studies which compared the newly implemented tool with the paper-based manual method in the first study (pilot study), and with the SQL Management Studio tool in the second study (full experiment). The results provided reasonable evidence that using SQL-FE can have a beneficial effect on formulating SQL statements and improve students' SQL learning. The results also showed that students were able to solve and formulate the SQL query on time and their performance showed significant improvement. The research also carried out an experiment to examine the viability of the SQL Marking Editor by testing the SQL partial marking, grouping of identical SQL statements, and the resulting marking process after applying the generic marking rules. The experimental results presented demonstrated that the newly implemented editor was able to provide consistent marking and individual feedback for all SQL parts. This means that the main aim of this research has been fulfilled, since the workload of the lecturers has been reduced, and students' performance in formulating SQL statements has been improved.

Although social media is a vital way to communicate and share knowledge, the researchers’ use is still less than expected. This may be due to their lack of self-efficacy or lack of knowledge of outcomes from its use. Therefore, this research aimed to investigate sources of self-efficacy and outcome expectations of researchers and their impact on the use of social media for knowledge sharing. To provide a theoretical framework, this research adopted social cognitive theory, which contains two important concepts are self-efficacy and outcome expectations. It investigated sources of self-efficacy and types of outcome expectations to address the research objectives and questions. This study has employed a sequential exploratory mixed methods design. It started with qualitative approach by conducting semi-structured interviews with thirty researchers from University of Strathclyde. The data were analysed by using a qualitative directed content analysis approach. In quantitative approach, online questionnaire was used to substantiate the qualitative findings. The total participants in this questionnaire was 144 researchers also from University of Strathclyde and the data were analysed by using descriptive statistics. This study found that researchers relied on the four sources of self-efficacy for using this media. They lead researchers to use it effectively, although some may discourage its use. It also found that researchers expect social and personal outcomes from its use. Each type has positive and negative forms, which can motivate or prevent researchers from use it. This study develops a theoretical framework by identifying levels of importance of these sources and types of outcomes as applied to a real-life online context. In a practical light, this helps researchers to understand these sources and outcomes and determine how they can develop in order to increase their confidence and use. Finally, institutions can encourage their staff, particularly researchers, to use it for their competitive advantage.

This research programme presented for consideration of the award of PhD by publication, presents a portfolio of 15 papers published between 2004 and 2018, and explores the transformation of public library digital services in a time of significant change. Initially exploring access issues from an organisational (policy) and architectural (design) perspective, ongoing work led to fundamental concerns over acceptable use from a user and ethical (behavioural) perspective. Significant access issues were initially discovered in relation to inconsistent digital service categorisation in architectural design, and restrictive policies and procedures. Further studies identified acceptable use policies that were not fit for purpose, and Internet filtering systems blocking legitimate content. Issues were also identified around the use of third party service providers, and the impact on user privacy that results, as well as what constituted acceptable use from the point of view of library patrons. Issues of public awareness of appropriate ethical behaviours then led to a critical examination of approaches to character education in information literacy education, notably identifying a lack of explicit attention to important aspects of ethical education in information literacy frameworks and models, and establishing a research agenda for further work. A range of research methods are utilised across the 15 papers that make up the submission, including literature reviews, surveys of library users and staff, heuristic evaluation of digital services, unobtrusive testing of access to public library services, content analysis of digital services, and content analysis of information literacy frameworks.

This thesis investigates if object oriented guidance is relevant in practice, and how this affects software that is produced. This is achieved by surveying practitioners and studying how constructs such as interfaces and inheritance are used in open-source systems. Surveyed practitioners framed 'good design' in terms of impact on development and maintenance. Recognition of quality requires practitioner judgement (individually and as a group), and principles are valued over rules. Time constraints heighten sensitivity to the rework cost of poor design decisions. Examination of open source systems highlights the use of interface and inheritance. There is some evidence of 'textbook' use of these structures, and much use is simple. Outliers are widespread indicating a pragmatic approach. Design is found to reflect the pressures of practice - high-level decisions justify 'designed' structures and architecture, while uncertainty leads to deferred design decisions - simpler structures, repetition, and unconsolidated design. Sub-populations of structures can be identified which may represent common trade-offs. Useful insights are gained into practitioner attitude to design guidance. Patterns of use and structure are identified which may aid in assessment and comprehension of object oriented systems.

Across the globe telecom firms are facing fierce competition amidst rapid technological advancements; thus, in order to succeed in the telecom business and remain profitable, satisfying and retaining loyal customers is vital. The process of providing superior services to customers who are socially interconnected has changed the landscape of customer relationship management (CRM). Following a detailed review of CRM, this study has identified the constructs that are critical for customer satisfaction in the Omani telecom sector. The literature review established that the primary reasons for organizations to implement CRM are to improve customer satisfaction, retain their customer base, obtain strategic information and enhance the value of the customer. It was also established that by collecting pertinent information on customers, such as their interests, habits and interactions, this enables organizations to provide a superior service which is specifically tailored to meet the individual needs of each customer. The sentiment analysis (SA) is particularly useful for CRM data collection. It is an efficient tool that has evolved through machine learning algorithms to provide an efficient and robust mechanism to observe customer sentiment in real-time. This research has demonstrated that such data will help to enhance CRM in Omani telecom firms because positive, negative and neutral sentiments alone cannot be used by firms to make strategic marketing decisions; therefore, sentiment analysis has been extended to include emotional analysis. To my knowledge, this is the first time that a novel visualisation tool has been used to demonstrate the choreographic emotion of tweets in real-time. With the recent push towards Marketing 3.0 where customers are considered to be multidimensional entities, emotional scores, more than sentiment scores, can play an active role in firms' decision-making processes. Questionnaires were developed for customers and social media managers based on the constructs identified during the literature review. Qualitative data was also collected through semi-structured interviews conducted with social media managers. Sentiment and emotional analysis were then applied to social media data pertaining to 83,981 tweets collected over one year in the Omani telecom sector. The data collected from social media managers, customers and findings of literature review were triangulated in order to arrive at meaningful conclusions. Based on this, a series of recommendations designed to enhance CRM in Omani telecom firms have been provided. Whilst 80% of the managers interviewed confirmed that they offer interactive online customer support, customers' responses relating to their level of satisfaction with the support they received remained neutral. This shows that the effectiveness of customer support remains questionable. Female customers indicated that they were satisfied with the support they received; however, male customers were not. Similarly, younger respondents (less than 19 years of age and between 20 to 30 years of age) responded with lower satisfaction scores than the other age groups. This information has to be taken into consideration during resolution management; the implication being that telecom firms should not only implement customer-centric CRM but also enable segment-focused service delivery. Students and government employees also provided lower scores for their satisfaction regarding the level of customer support they received. I have identified that the emotional scores of live social media data can play a big role in CRM processes, and when these data were visualised as a choreography of emotions using Plutchik's Wheel of Emotion, it was observed that the customers who received better customer service may have experienced feelings such as 'surprise' and 'happiness'. Conversely, when customers were not satisfied, the emotions shown included 'disappointment', 'anger', 'irritation' and/or 'anxiety'. These findings have been mapped on the Wheel of Emotion. This study, therefore, led to developing a model that integrates the SA of social media data along with the CRM processes of telecom firms in real-time. This involves horizontal and vertical integration of the outcomes of emotional analyses to be shared across the organisation. I recommended the implementation of a 7-step approach to capitalise on the robustness of machine learning algorithms by choreographing emotions from live stream social media data. This can equip firms with better tools as they strive to prosper in a world of competitors and rapidly advancing technology.

Microservices have gained wide recognition and acceptance in software industries as an emerging architectural style for autonomic, scalable, and more reliable computing. In this thesis we target a critical, main problem related to the transition towards microservices: reasoning about the suitable granularity level of a microservice (i.e. when and how to merge or decompose microservices). We conduct a systematic mapping study and use it to identify inadequacies in the state-of-the-practice and -art related to this problem. The thesis addresses the following inadequacies: a relatively disciplined understanding of the transition to microservices and technical activities underlying it, systematic architecture-oriented modelling support for microservice granularity, a dynamic architectural evaluation process to reason about the cost and added value of granularity adaptation, and effective decision support to inform reasoning about microservice granularity at runtime. To address the identified inadequacies, initially we contribute an architecture-centric modelling approach for microservices. Next, we contribute a dynamic evaluation process which links granularity adaptation to its added value under uncertainty. Next, we contribute an interactive, iterative planning engine to provide insight regarding which granularity adaptation strategy is suitable at runtime. We use a hypothetical microservice application - Filmflix - as a case study for evaluating each aforementioned contribution. Finally, this thesis contributes to microservice-specific guidance aiming to render scalability-aware granularity adaptation decisions. We evaluate this contribution by comparing its usage against ad-hoc scalability analysis.

Though investigated for decades, Software Effort Estimation (SEE) remains a challenging problem in software project management. However, there are several factors hindering the practical use of SEE models. One major factor is the scarcity of software projects that are used to construct SEE models due to the long process of software development. Even given a large number of projects, the collected effort values are usually corrupted by noise due to the participation of humans. Furthermore, even given enough and noise-free software projects, SEE models may have sensitive parameters to tune possibly causing model sensitivity problem. The thesis focuses on tackling these three issues. It proposes a synthetic data generator to tackle the data scarcity problem, introduces/constructs uncertain effort estimators to tackle the data noise problem, and analyses the sensitivity to parameter settings of popular SEE models. The main contributions of the thesis include: 1. Propose a synthetic project generator and provide an understanding of when and why it improves prediction performance of what baseline models. 2. Introduce relevance vector machine for uncertain effort estimation. 3. Propose a better uncertain estimation method based on an ensemble strategy. 4. Provide a better understanding of the impact of parameter tuning for SEE methods.

Automatic classification of complex data is an area of great interest as it allows to make efficient use of the increasingly data intensive environment that characterizes our modern world. This thesis presents to two contributions to this research area. Firstly, the problem of discriminative feature extraction for data organized in multidimensional arrays. In machine learning, Linear Discriminant Analysis (LDA) is a popular discriminative feature extraction method based on optimizing a Fisher type criterion to find the most discriminative data projection. Various extension of LDA to high-order tensor data have been developed. The method proposed is called the Efficient Greedy Feature Extraction method (EGFE). This method avoids solving optimization problems of very high dimension. Also, it can be stopped when the extracted features are deemed to be sufficient for a proper discrimination of the classes. Secondly, an application of EGFE methods to early detection of dementia disease. For the early detection task, four cognitive scores are used as the original data while we employ our greedy feature extraction method to derive discriminative privileged information feature from fMRI data. The results from the experiments presented in this thesis demonstrate the advantage of using privileged information for the early detection task.

Agents operating in domains with time budgets shared between planning and execution must carefully balance the need to plan versus the need to act. This is because planning and execution consume the same time resource. Excessive planning can delay the time it takes to achieve a goal, and so reduce the reward attained by an agent. Whereas, insufficient planning will mean the agent creates and executes low reward plans. This thesis looks at three ways to increase the reward achieved by an agent in domains with shared time budgets. The first way is by optimising time allocated to planning, using two different methods -- an optimal plan duration predictor and an online loss limiter. A second is by finding ways to act in a goal-directed manner during planning. We look at using previous plans or new plans generated quickly as heuristics for acting whilst planning. In addition, we present a way of describing actions that are mid-execution to speed the transition between planning and execution. Lastly, this thesis presents a way in which to manage time budgets in multi-agent domains. We use market-based task allocation with deadlines to produce faster task allocation and planning.

Visual search is a cognitive process that makes use of eye movements to bring the relatively high acuity fovea to bear on areas of interest to aid in navigation or interaction within the environment. This thesis explores a novel hypothesis that human visual search behaviour emerges as an adaptation to the underlying human information processing constraint, task utility and ecology. A new computational model (Computationally Rational Visual Search (CRVS) model) for visual search is also presented that provides a mathematical formulation for the hypothesis. Through the model, we ask the question, what mechanism and strategy a rational agent would use to move gaze and when should it stop searching? The CRVS model formulates the novel hypothesis for visual search as a Partially Observable Markov Decision Process (POMDP). The POMDP provides a mathematical framework to model visual search as a optimal adaptation to both top-down and bottom-up mechanisms. Specifically, the agent is only able to partially observe the environment due to the bounds imposed by the human visual system. The agent learns to make a decision based on the partial information it obtained and a feedback signal. The POMDP formulation is very general and it can be applied to a range of problems. However, finding an optimal solution to a POMDP is computationally expensive. In this thesis, we use machine learning to find an approximately optimal solution to the POMDP. Specifically, we use a deep reinforcement learning (Asynchronous Advantage Actor-Critic) algorithm to solve the POMDP. The thesis answers the where to fixate next and when to stop search questions using three different visual search tasks. In Chapter 4 we investigate the computationally rational strategies for when to stop search using a real-world search task of images on a web page. In Chapter 5, we investigate computationally rational strategies for where to look next when guided by low-level feature cues like colour, shape, size. Finally, in Chapter 6, we combine the approximately optimal strategies learned from the previous chapters for a conjunctive visual search task (Distractor-Ratio task) where the model needs to answer both when to stop and where to search question. The results show that visual search strategies can be explained as an approximately optimal adaptation to the theory of information processing constraints, utility and ecology of the task.

The European Rail Traffic Management System (ERTMS) is a new signalling scheme that is being implemented worldwide with the aim of improving interoperability and cross-border operation. It is also an example of an Industrial Control System, a safety-critical system which, in recent years, has been subject to a number of attacks and threats. In these systems, safety is the primary concern of the system designers, whilst security is sometimes an afterthought. It is therefore prudent to assure the security for current and future threats, which could affect the safe operation of the railway. In this thesis, we present a systematic security analysis of parts of the ERTMS standard, firstly reviewing the security offered by the protocols used in ERTMS using the ProVerif tool. We will then assess the custom MAC algorithm used by the platform and identify issues that exist in each of the ERTMS protocol layers, and aim to propose solutions to those issues. We also identify a challenge presented by the introduction of ERTMS to National Infrastructure Managers surrounding key management, where we also propose a novel key management scheme, TRAKS, which reduces its complexity. We then define a holistic process for asset owners to carry out their own security assessments for their architectures and consider the unique challenges that are presented by Industrial Control Systems and how these can be mitigated to ensure security of these systems. Drawing conclusions from these analyses, we introduce the notion of a `secure architecture' and review the current compliance of ERTMS against this definition, identifying the changes required in order for it to have a secure architecture, both now and also in the future.

Context: Design-time evaluation is essential to build the initial software architecture to be deployed. However, experts' design-time assumptions are unlikely to remain true indefinitely in systems characterized by scale, heterogeneity, and dynamism (e.g. IoT). Experts' design-time decisions can be thus challenged at run-time. A continuous architecture evaluation that systematically intertwines design-time and run-time evaluation is necessary. However, the literature lacks examples on how continuous evaluation can be realized and conducted. Objective: This thesis proposes the first continuous architecture evaluation framework. Method: The framework is composed of two phases: design-time and run-time evaluation. The design-time evaluation enables the necessary initial step of system design and deployment. Run-time evaluation assesses to what extent the architecture options adopted at design-time and other potential options, perform well at run-time. For that, the framework leverages techniques inspired by finance, reinforcement learning, multi-objective optimisation, and time series forecasting. The framework can actively track and proactively forecast the performance of architecture decisions and detect any detrimental changes. It can then inform deployment, refinement, and/or phasing-out decisions. We use an IoT case study to show how continuous evaluation can fundamentally guide the architect and influence the outcome of the decisions. A series of experiments is conducted to demonstrate the applicability and effectiveness of the framework. Results: The design-time evaluation was able to evaluate the architecture options under uncertainty and shortlist candidates for further refinement at run-time. The run-time evaluation has shown to be effective. In particular, it enabled a significant improvement in overall quality (about 40-70% better than reactive and state-of-the-art approaches in some scenarios), with enhanced architecture's stability. It was also shown to be scalable and robust to various noise levels. In addition, it provides the architect with flexibility to set a monitoring interval to profile the quality of candidates and has parameters that enable the architect to manage the trade-off between architecture stability and learning accuracy. Conclusion: The proposed continuous evaluation framework could potentially aid the architect in evaluating complex design decisions in dynamic environments.

This thesis develops practical Bayesian estimators and exploration methods for count data collected by autonomous robots with unreliable sensors for long periods of time. It addresses the problems of drawing inferences from temporally incomplete and unreliable count data. This thesis contributes statistical models with spectral analysis which are able to capture the periodic structure of count data on extended temporal scales from temporally sparse observations. It is shown how to use these patterns to i) predict the human activity level at particular times and places and ii) categorize locations based on their periodic patterns. The second main contribution is a set of inference methods for a Poisson process which takes into account the unreliability of the detection algorithms used to count events. Two tractable approximations to the posterior of such Poisson processes are presented to cope with the absence of a conjugate density. Variations of these processes are presented, in which (i) sensors are uncorrelated, (ii) sensors are correlated, (iii) the unreliability of the observation model, when built from data, is accounted for. A simulation study shows that these partially observable Poisson process (POPP) filters correct the over- and under-counts produced by sensors. The third main contribution is a set of exploration methods which brings together the spectral models and the POPP filters to drive exploration by a mobile robot for a series of nine-week deployments. This leads to (i) a labelled data set and (ii) solving an exploration exploitation trade-off: the robot must explore to find out where activities congregate, so as to then exploit that by observing as many activities.

Elasticity is the characteristic of cloud computing that provides the underlying primitives to dynamically acquire and release shared computational resources on demand. Moreover, it unfolds the advantage of the economies of scale in the cloud, which refers to a drop in the average costs of these computing capacities as a result of the dynamic sharing capability. However, in practice, it is impossible to achieve elasticity adaptations that obtain perfect matches between resource supply and demand, which produces dynamic gaps at runtime. Moreover, elasticity is only a capability, and consequently it calls for a management process with far-sighted economics objectives to maximise the value of elasticity adaptations. Within this context, we advocate the use of an economics-driven approach to guide elasticity managerial decisions. We draw inspiration from the technical debt metaphor in software engineering and we explore it in a dynamic setting to present a debt-aware elasticity management. In particular, we introduce a managerial approach that assesses the value of elasticity decisions to adapt the resource provisioning. Additionally, the approach pursues strategic decisions that value the potential utility produced by the unavoidable gaps between the ideal and actual resource provisioning over time. As part of experimentation, we built a proof of concept and the results indicate that value-oriented adaptations in elasticity management lead to a better economics performance in terms of lower operating costs and higher quality of service over time. This thesis contributes (i) an economics-driven approach towards elasticity management; (ii) a technical debt-aware model to reason about elasticity adaptations; (iii) a debt-aware learning elasticity management approach; and (iv) a multi-agent elasticity management for multi-tenant applications hosted in the cloud.

Rule-based expert systems (RBSs) provide an efficient solution to many problems that involve event stream processing. With today's needs to process larger streams, many approaches have been proposed to distribute the rule engines behind RBSs. However, there are some issues which limit the potential of distributed RBSs in the current big data era, such as the load imbalance due to their distribution methods, and low parallelism originated from the continuous operator model. To address these issues, we propose a new architecture for distributing rule engines. This architecture adopts the dynamic job assignment and the micro-batching strategies, which have recently arisen in the big data community, to remove the load imbalance and increase parallelism of distributed rule engines. An automated transformation framework based on Model-driven Architecture (MDA) is presented, which can be used to transform the current rule engines to work on the proposed architecture. This work is validated by a 2-step verification. In addition, we propose a generic benchmark for evaluating the performance of distributed rule engines. The performance of the proposed architecture is discussed and directions for future research are suggested. The contribution of this study can be viewed from two different angles: for the rule-based system community, this thesis documents an improvement to the rule engines by fully adopting big data technologies; for the big data community, it is an early proposal to process large event streams using a well crafted rule-based system. Our results show the proposed approach can benefit both research communities.

Gender imbalance in STEM disciplines (Science, Technology, Engineering and Mathematics) has been a research subject for years. Studies have shown potential reasons leading to the underrepresentation of women in such disciplines and have suggested why and how to improve the gender balance and women's experiences in these areas. The High Performance Computing (HPC) community, which spans various STEM subjects and relies on advanced scientific research, might present a similar picture. The aim of this thesis is to understand the gender demographics of the HPC community, to identify the underlying reasons of a potential gender imbalance, and to suggest effective ways of improvement. Since HPC is such a broad community, to obtain a first picture of the proportion of women in the HPC community, we decided to examine historical demographics of two different settings which are potential indicators of the participation and contribution of women in the community, namely HPC-related conferences and HPC training courses. From the analysis of these quantitative data, we found that women were fewer than men in all the categories of conference participation that we examined, and that women were outnumbered by men at all levels and years of the courses examined. Our study reveals an underrepresentation of women in the HPC community, along the lines of what already observed in STEM disciplines. Additionally, we conducted a survey in order to further understand the reasons of the gender imbalance and to find out from the people within the HPC community what could be done to address the issue. Results from our survey indicate that the clear majority of both women and men forming the HPC community come from a STEM background, which is considered as the main reason of women's underrepresentation by the participants of this study. We also discovered that women are less likely to receive training and to develop software, both crucial factors for using HPC facilities for research purposes. Gender differences are also found in the impact of parenthood on career progression; the perception of gender discrimination in workplace and conference environments; the importance of gender balance, mentoring, role models and Equality and Diversity awareness in the HPC community. Similar findings and gender differences are also highlighted and confirmed by the results of further qualitative approaches of this study. We conducted interviews and focus group discussions with selected and recommended individuals of the community, to support and interpret previously obtained data, and to stimulate new ideas or hypotheses for future work. According to the interviewees and the participants to focus group discussions, one of the main challenges of the HPC community is its image of a closed, inaccessible, "geeky" area, which focuses on the size, speed and power of supercomputers, rather than on their use for solving problems in research and in life. This might be one reason that makes the community unattractive to women. Also, of significant importance for the current diversity status of the community is the fact that HPC is not well-promoted as a research tool, especially to more gender-balanced non-STEM subjects, in combination with the lack of formal (HPC and programming) training and of women in senior positions. This thesis forms the first step to understand the womens representation and experiences within the HPC community. All the topics studied, and the evidence gathered in this thesis have provided significant insight to enable further research on the best practices for improvement in the HPC community and related STEM fields.

In this Thesis we use colour and Local Binary Pattern based texture analysis for image classification and reconstruction. In complementary work we offer a new texture description called the Sudoku transform, an extension of the Local Binary Pattern. Our new method when used to classify members of benchmark datasets shows a performance increment over traditional methods including the Local Binary Pattern. Finally we consider the invertibility of texture descriptions and show how with our new method - Quadratic Reconstruction - that a highly accurate image can be recovered purely from its textural information.

Biomolecular interactions underpin many of the processes that make up life. Molecular docking is the study of these interactions in silico. Interactive docking applications put the user in control of the docking process, allowing them to use their knowledge and intuition to determine how molecules bind together. Interactive molecular docking applications often use haptic devices as a method of controlling the docking process. These devices allow the user to easily manipulate the structures in 3D space, whilst feeling the forces that occur in response to their manipulations. As a result of the force refresh rate requirements of haptic devices, haptic assisted docking applications are often limited, in that they model the interacting proteins as rigid, use low fidelity visualisations or require expensive propriety equipment to use. The research in this thesis aims to address some of these limitations. Firstly, the development of a visualisation algorithm capable of rendering a depiction of a deforming protein at an interactive refresh rate, with per-pixel shadows and ambient occlusion, is discussed. Then, a novel approach to modelling molecular flexibility whilst maintaining a stable haptic refresh rate is developed. Together these algorithms are presented within Haptimol FlexiDock, the first haptic-assisted molecular docking application to support receptor flexibility with high fidelity graphics, whilst also maintaining interactive refresh rates on both the haptic device and visual display. Using Haptimol FlexiDock, docking experiments were performed between two protein-ligand pairs: Maltodextrin Binding Protein and Maltose, and glutamine Binding Protein and Glucose. When the ligand was placed in its approximate binding site, the direction of over 80% of the intra-molecular movement aligned with that seen in the experimental structures. Furthermore, over 50% of the expected backbone motion was present in the structures generated with FlexiDock. Calculating the deformation of a biomolecule in real time, whilst maintaining an interactive refresh rate on the haptic device (> 500Hz) is a breakthrough in the field of interactive molecular docking, as, previous approaches either model protein flexibility, but fail to achieve the required haptic refresh rate, or do not consider biomolecular flexibility at all.