Search

Machine learning (ML) and artificial intelligence (AI) are rapidly growing fields with applications in many scientific domains. In this work we focus on supervised learning algorithms for prediction tasks in practical applications. Specifically, we focus on two applications, predicting adverse events in low-carbon energy production and screening patients' eligibility for implantable defibrillators. We present optimised pre-processing methods, optimise hyperparameter selections and present our own bilevel optimisation model for simultaneous training and hyperparameter tuning. We first consider the problem of predicting infrequently occurring adverse events from time series data (Provided by Andigestion Ltd, a UK-based anaerobic digestion company, and a civil nuclear power plant in the UK) for which we propose a framework for modelling this problem as an imbalanced classification task and we construct and compare numerous models and sampling techniques. The models developed here could be integrated into a decision support tool for providing advanced warning of adverse events which can lead to significant commercial benefits. We then propose an AI tool for automated, prolonged screening of patients' implantable defibrillator eligibility which is created using real ECG data provided by our partners at the University Hospital Southampton. As we demonstrate in our experiments, this tool is capable of predicting patients' T:R ratios (a major indication of implantation eligibility) to within 0.0461 of their true values. This level of accuracy is sufficient to facilitate the automation of the measurement process. We show how this tool can enable cardiologists to perform 24-hour automated screenings, thus allowing them to better determine patients' eligibility for implantation. Finally, we formulate the bilevel problem of hyperparameter selection for non-linear kernel support vector machines via k-fold cross validation and propose an algorithm for solving it, for which we demonstrate some convergence properties. We provide a number of examples of this algorithm in use on a number of real data sets from the UCI repository.

This thesis examines a photographic archive created by John Laing and Son Ltd, a major UK civil engineering and housebuilding company, to improve our understanding of the documentation of Laing's construction process in the UK and the company which undertook the work. The thesis starts by developing a history of Laing, describing how the John Laing Photographic Collection amplifies works which were important to the company and offers a perspective on construction products and processes employed. It demonstrates how the retention of images drives a specific and intended presentation of the company. The second theme debates the use of photography in the development of Laing's internal and external corporate identities. The use of photography in advertising and as a component of internal communications is considered. The final theme explored is the presentation of the built environment through the construction phase. It is argued that construction photography has a different character to that of architectural photography. This difference in character is evidenced through examination of a small number of specific works undertaken by Laing. The practice and profession of construction photography is compared to that of architectural photographers. The thesis concludes by arguing that while a single photograph captures a moment in time, the John Laing Photographic Collection presents a flow of events that can be viewed as a process of physical and immaterial creation through the construction of buildings and identity. This sense of a process is intrinsic to construction and is proposed as a differentiator to much of what is shown in architectural photography with its emphasis on a presentation of a completed structure, often frozen at the conclusion of construction and before habitation. Implicitly the thesis demonstrates the value of photography to historians of the built environment and especially the process of construction.

Sir William is a literary novel based on the true story of the celebrated Kent lunatic, Sir William Courtenay, who led a rural rebellion in Hernhill in 1838, culminating in what has been dubbed 'the last battle fought on English soil'. If you can call over 100 well-armed soldiers facing 30 farm labourers armed only with sticks and farm implements a battle? The outcome was as one might expect. The story is seen from the point of view of Adman Packman, an unfulfilled ex-soldier, who gets caught up in Sir William's seductive personality cult. It touches on contemporary themes about the ignored and dissatisfied which we can see in the followers of popular politics, from Trump to Brexit to Isis. And in common with those it mixes a form of liberation politics with religious fundamentalism and messianic dreams - the need to have something, no matter how crazy, to believe in.

A strong sense of professional identity is vital for a teacher's understanding, engagement and commitment to their role and helps them make sense of their work and wider relationships. It is also a key element in the development of agency and career stability. Understanding teacher professional identity helps policymakers make sense of how teachers commit to, and engage with, policy making formulation and enhances implementation. This research study ascertains the main characteristics of teacher professional identity through a review of the existing literature. Guided by Michel Foucault's theory of power, it uses critical discourse analysis of 32 policy texts to uncover the way that government policy discourse may shape this identity. Through semistructured interviews with 12 practising teachers, the study evaluates the extent to which policy discourse shapes teacher identity of secondary school teachers in England. Findings suggest four characteristics of teacher professional identity: the self; knowledge of the subject discipline and how to engage pupils in this; the impact of experience; and the centrality of emotions, both in seeing teaching as a vocation and engaging student emotions. The critical discourse analysis indicates government policy discourse presents teachers in a negative light in a consistent and sustained manner. Far from being afforded professional status, teaching is presented as a technical skill with teachers charged with inadequacy and failing pupils through an unthinking enthralment to progressive pedagogies and an unwillingness to reform. Greater accountability within a reformed autonomous corporate structure with teachers taking more responsibility for social and economic policy are central planks of this reform agenda. Interviews with teachers suggest that teacher professional identity is indeed shaped by government policy discourse, but not in the way policymakers may hope. Teachers consistently push back on claims of inadequacy and embrace their professional status. Policymakers' claims of the dominance of progressive ideologies is dismissed by a pragmatic approach to classroom practice. Far from shaping the teaching profession, policy discourse highlights deep misunderstandings between the profession and policymakers entrenching teacher identity further away from policy objectives.

Pharmaceuticals are most commonly studied in randomised controlled trials (RCTs) against a control arm (either active, or placebo). On occasions however treatments are licensed exclusively on the basis of uncontrolled study data - this thesis investigates how comparative effectiveness can be estimated under such circumstances. The role of RCTs in the approval and estimation of comparative effectiveness in pharmaceuticals is discussed, as well as potential methods for analysis where RCT data are not available. A review of all drug approvals from 1999-2014 by the European Medicines Agency and the US Food and Drug Administration is then presented. Performing literature searches in the majority of cases (80%), historical controls have been the primary source of estimates of comparative effectiveness, frequently without attempts to adjust for differences between studies. Given the high usage of historical controls, I focussed on the role of adjustment. This included a simulation study to understand where the method of Matching Adjusted Indirect Comparison (MAIC) is likely to be of use, looking specifically at the effect of model misspecification. Three novel methods (with practical examples) for the creation of historical controls are then presented; using extrapolation from the previous line of therapy, using non-responders to therapy as a surrogate, and comparing to a patient’s own prior data. The conclusion of the work is that there are clearly situations where RCTs cannot, or will not be used – regardless of the statistical issues this raises. In such cases by proactively identifying appropriate historical data, and using appropriate analysis methods – the downsides can be ameliorated, at least in part. A flowchart presenting the available methods (split by data access) is presented. Further research is required on the appropriateness of different sources of historical control data (e.g. registries versus RCT arms), and how to synthesize multiple estimates of effectiveness (e.g. multiple MAICs).

It is well known that at thermal equilibrium (whereby a system has settled into a steady state with no energy or mass being exchanged with the environment), the microstates of a system are exponentially weighted by their energies, giving a Boltzmann distribution. All macroscopic quantities, such as the free energy and entropy, can be in principle computed given knowledge of the partition function. In a nonequilibrium steady state, on the other hand, the system has settled into a stationary state, but some currents of heat or mass persist. In the presence of these currents, there is no unified approach to solve for the microstate distribution. This motivates the central theme of this work, where I frame and solve problems in nonequilibrium statistical physics in terms of random walk and diffusion problems. The system that is the focus of Chapters 2, 3, and 4 is the (Totally) Asymmetric Simple Exclusion Process, or (T)ASEP. This is a system of hard-core particles making jumps through an open, one-dimensional lattice. This is a paradigmatic example of a nonequilibrium steady state that exhibits phase transitions. Furthermore, the probability of an arbitrary configuration of particles is exactly calculable, by a matrix product formalism that lends a natural association between the ASEP and a family of random walk problems. In Chapter 2 I present a unified description of the various combinatorial interpretations and mappings of steady-state configurations of the ASEP. As well as deriving new results, I bring together and unify results and observations that have otherwise been scattered in the combinatorics and physics literature. I show that particular particle configurations of the ASEP have a one-to-many mapping to a set of more abstract paths, which themselves have a one-to-many mapping to permutations of numbers. One observation from this wider literature has been that this mapped space can be interpreted as a larger set of configurations in some equilibrium system. This naturally gives an interpretation of ASEP configuration probabilities as summations of Boltzmann weights. The nonequilibrium partition function of the ASEP is then a summation over this equilibrium ensemble, however one encounters difficulties when calculating more detailed measures of this state space, such as the entropy. This motivates the work in Chapter 3. I calculate a quantity known as the Rényi entropy, which is a measure of the partitioning of the state space, and a deformation of the familiar Shannon entropy. The Rényi entropy is simple for an equilibrium system, but has yet to be explored in a classical nonequilibrium steady state. I use insights from Chapter 2 to frame one of these Rényi entropies | requiring the enumeration of the squares of configuration weights | in terms of a two-dimensional random walk with absorbing boundaries. I find the appropriate generating function across the full phase diagram of the TASEP by generalising a mathematical technique known as the obstinate kernel method. Importantly, this nonequilibrium Rényi entropy has a different structural form to any equilibrium system, highlighting a clear distinction between equilibrium and nonequilibrium distributions. In Chapter 4 I continue to examine the Rényi entropy of the TASEP, but now performing a time and space continuum limit of the random walk problem in Chapter 3. The resultant problem is a two-dimensional dffusion problem with absorbing boundary conditions, which once solved should recover TASEP dynamics about the point in the phase diagram where the three dynamical phases meet. I derive a generating function, sufficiently simple that its singularities can be analysed by hand. This calculation entails a novel generalisation of the obstinate kernel method of Chapter 3: I find a solution by exploiting a symmetry in the Laplace transform of the diffusion equation. I finish in Chapter 5 by introducing and solving another nonequilibrium system, termed the many-filament Brownian ratchet. This comprises an arbitrary number of filaments that stochastically grow and contract, with the net effect of moving a drift-diffusing membrane by purely from thermal fluctuations and steric interactions. These dynamics draw parallels with those of actin filament networks at the leading edge of eukaryotic cells, and this improves on previous 'pure ratchet' models by introducing interactions and heterogeneity in the filaments. I find an N-dimensional diffusion equation for the evolution of the N filament-membrane displacements. Several parameters can be varied in this system: the drift and diffusion rates of each of the filaments and membrane, the strength of a quadratic interaction between each filament with the membrane, and the strength of a surface tension across the filaments. For several interesting physical cases I find the steady-state distribution exactly, and calculate how the mean velocity of the membrane varies as a function of these parameters.

Current methods of clinical guideline development have two large challenges: 1) there is often a long time-lag between the key results and publication into recommended best practice and 2) the measurement of adherence to those guidelines is often qualitative and difficult to standardise into measurable impact. In an age of ever-increasing volumes of accurate data captured at the bedside in specialist intensive care units, this thesis explores the possibility of constructing a technology that can interpret that data and present the results as a quantitative and immediate measure of guideline adherence. Applied to the Traumatic Brain Injury (TBI) domain, and specifically to the management of ICP and CPP, a framework is developed that makes use of process models to measure the adherence of clinicians to three specific TBI guidelines. By combining models constructed from physiological and treatment ICU data, and those constructed from guideline text, a distance is calculated between the two, and patterns of guideline adherence are inferred from this distance. The framework has been developed into an online application capable of producing adherence output on most standardised ICU datasets. This application has been applied to the Brain-IT and MIMIC III repositories and evaluated on the Philips ICCA bedside monitoring system. Patterns of guideline adherence are presented in a variety of ways including minute-by-minute windowing, tables of non-adherence instances, statistical distribution of instances, and a severity chart summarising the impact of non-adherence in a single number.