Ammonia plays a central role in the pathogenesis of cerebral oedema in paracetamol-induced acute liver failure (PALF). Infection and inflammation play an important synergistic role in its development. Toll-like receptors (TLRs) sense pathogens and induce inflammation but whether this contributes to the development of cerebral oedema in PALF remains unknown. I postulated that ammonia-induced cerebral oedema and immune dysfunction are mediated by TLR9 and aimed to determine whether this could be prevented in a hyperammonemic TLR9 knockout mouse model. TLR9 expression on circulating neutrophils and their function in PALF was assessed. To examine the influence of PALF plasma and endogenous DNA on TLR9 expression, healthy neutrophils were incubated with PALF plasma with/without DNase. Ammonium acetate (NH4-Ac) was injected intraperitoneally in wild type Black6 (WT-B6), TLR9-/- B6 mice and TLR9fl/fl LysCre B6 mice with TLR9 deleted from neutrophils and macrophages. The TLR9 antagonist ODN2088 was also evaluated. Neutrophil TLR9 correlated with plasma IL-8 and ammonia concentration and increased with severity of hepatic encephalopathy and systemic inflammation. Healthy neutrophil TLR9 expression increased upon stimulation with PALF plasma which was abrogated by pre-incubation with DNase. Following NH4-Ac stimulation, intracellular cytokine (IFN-γ, TNF-α and IL-6) production of lymphocytes and macrophages were increased in WT-B6 mice compared to controls. This was accompanied by increased brain water however in TLR9-/-, cytokine production and brain water content were decreased. This was seen similarly in WT-B6 administered the TLR9 antagonist ODN2088 in conjunction with NH4-Ac. TLR9fl/fl LysCre mice had decreased cytokine production and brain water compared to the TLR9fl/fl group following NH4-Ac injection. Total DNA levels were increased in the circulation after NH4-Ac injection. In summary, ammonia-induced cerebral oedema and immune dysfunction are mediated through TLR9 and DNA dependent. The amelioration of brain oedema and lymphocyte cytokine production by ODN2088 supports exploration of TLR9 antagonism in early PALF to prevent progression to cerebral oedema.

The discipline of biometrics may be broadly defined as the study of using metrics related to human characteristics as a basis for individual identification and authentication, and many approaches have been implemented in recent years for many different scenarios. A sub-section of biometrics, specifically known as soft biometrics, has also been developing rapidly, which focuses on the additional use of information which is characteristic of a user but not unique to one person, examples including subject age or gender. Other than its established value in identification and authentication tasks, such useful user information can also be predicted within soft biometrics modalities. Furthermore, some most recent investigations have demonstrated a demand for utilising these biometric modalities to extract even higher-level user information, such as a subject\textsc{\char13}s mental or emotional state. The study reported in this thesis will focus on investigating two soft biometrics modalities, namely keystroke dynamics and handwriting biometrics (both examples of hand-based biometrics, but with differing characteristics). The study primarily investigates the extent to which these modalities can be used to predict human emotions. A rigorously designed data capture protocol is described and a large and entirely new database is thereby collected, significantly expanding the scale of the databases available for this type of study compared to those reported in the literature. A systematic study of the predictive performance achievable using the data acquired is presented. The core analysis of this study, which is to further explore of the predictive capability of both handwriting and keystroke data, confirm that both modalities have the capability for predicting higher level mental states of individuals. This study also presents the implementation of detailed experiments to investigate in detail some key issues (such as amount of data available, availability of different feature types, and the way ground truth labelling is established) which can enhance the robustness of this higher level state prediction technique.

In this thesis, we address problems encountered in complex network analysis using graph theoretic methods. The thesis specifically centers on the challenge of how to characterize the structural properties and time evolution of graphs. We commence by providing a brief roadmap for our research in Chapter 1, followed by a review of the relevant research literature in Chapter 2. The remainder of the thesis is structured as follows. In Chapter 3, we focus on the graph entropic characterizations and explore whether the von Neumann entropy recently defined only on undirected graphs, can be extended to the domain of directed graphs. The substantial contribution involves a simplified form of the entropy which can be expressed in terms of simple graph statistics, such as graph size and vertex in-degree and out-degree. Chapter 4 further investigates the uses and applications of the von Neumann entropy in order to solve a number of network analysis and machine learning problems. The contribution in this chapter includes an entropic edge assortativity measure and an entropic graph embedding method, which are developed for both undirected and directed graphs. The next part of the thesis analyzes the time-evolving complex networks using physical and information theoretic approaches. In particular, Chapter 5 provides a thermodynamic framework for handling dynamic graphs using ideas from algebraic graph theory and statistical mechanics. This allows us to derive expressions for a number of thermodynamic functions, including energy, entropy and temperature, which are shown to be efficient in identifying abrupt structural changes and phase transitions in real-world dynamical systems. Chapter 6 develops a novel method for constructing a generative model to analyze the structure of labeled data, which provides a number of novel directions to the study of graph time-series. Finally, in Chapter 7, we provide concluding remarks and discuss the limitations of our methodologies, and point out possible future research directions.

This thesis is concerned with embedded systems which employ time-triggered software architectures and for which there are both severe resource constraints and a requirement for highly-predictable behaviour. The thesis discusses design patterns and their benefits to software development and reviews a pattern language (the PTTES collection) previously assembled to support the development of time-triggered embedded systems. As embedded systems become ever more complex and - in many cases - take on an increasing role in safety, it is widely recognised that developers require tools and techniques that support the 'automatic' generation of such designs. This thesis makes a novel contribution to the field of pattern-based automated code generation and illustrates the capabilities of this approach in the development of reliable time-triggered embedded systems. Specifically, the approach described in this thesis addresses a key limitation of previous work in this area, namely the challenge of implementing the 'one pattern, many implementations' relationship. Furthermore, unlike previous pattern tools, the approach described in this thesis is based on a substantial pattern language: this paves the way for the generation of coherent application code from groups of related patterns. To test the above ideas, the thesis describes PTTES Builder, a pattern-based code generation tool based on the PTTES collection. In an empirical study, the effectiveness of the PTTES Builder approach is compared with an equivalent 'manual' approach. The results obtained demonstrate that time-triggered embedded systems can be created using this approach. There is also some evidence that the use of the tool is likely to lead to improved code reliability and quality. In a second study discussed in the thesis, there are indications that the approach implemented by PTTES Builder is robust enough to support the evolution of its underlying pattern collection. The thesis concludes by making a number of suggestions for future extensions to this work.

The need for automatic recognition and understanding of speech is emerging in tasks involving the processing of large volumes of natural conversations. In application domains such as Search and Rescue, exploiting automated systems for extracting mission-critical information from speech communications has the potential to make a real difference. Spoken language understanding has commonly been approached by identifying units of meaning (such as sentences, named entities, and dialogue acts) for providing a basis for further discourse analysis. However, this fine-grained identification of fundamental units of meaning is sensitive to high error rates in the automatic transcription of noisy speech. This thesis demonstrates that topic segmentation and identification techniques can be employed for information extraction from spoken conversations by being robust to such errors. Two novel topic-based approaches are presented for extracting situational information within the search and rescue context. The first approach shows that identifying the changes in the context and content of first responders' report over time can provide an estimation of their location. The second approach presents a speech-based topological map estimation technique that is inspired, in part, by automatic mapping algorithms commonly used in robotics. The proposed approaches are evaluated on a goal-oriented conversational speech corpus, which has been designed and collected based on an abstract communication model between a first responder and a task leader during a search process. Results have confirmed that a highly imperfect transcription of noisy speech has limited impact on the information extraction performance compared with that obtained on the transcription of clean speech data. This thesis also shows that speech recognition accuracy can benefit from rescoring its initial transcription hypotheses based on the derived high-level location information. A new two-pass speech decoding architecture is presented. In this architecture, the location estimation from a first decoding pass is used to dynamically adapt a general language model which is used for rescoring the initial recognition hypotheses. This decoding strategy has resulted in a statistically significant gain in the recognition accuracy of the spoken conversations in high background noise. It is concluded that the techniques developed in this thesis can be extended to more application domains that deal with large volumes of natural spoken conversations.

Many electronic devices are beginning to include Voice User Interfaces (VUIs) as an alternative to conventional interfaces. VUIs are especially useful for users with restricted upper limb mobility, because they cannot use keyboards and mice. These users, however, often suffer from speech disorders (e.g. dysarthria), making Automatic Speech Recognition (ASR) challenging, thus degrading the performance of the VUI. Partially Observable Markov Decision Process (POMDP) based Dialogue Management (DM) has been shown to improve the interaction performance in challenging ASR environments, but most of the research in this area has focused on Spoken Dialogue Systems (SDSs) developed to provide information, where the users interact with the system only a few times. In contrast, most VUIs are likely to be used by a single speaker over a long period of time, but very little research has been carried out on adaptation of DM models to specific speakers. This thesis explores methods to adapt DM models (in particular dialogue state tracking models and policy models) to a specific user during a longitudinal interaction. The main differences between personalised VUIs and typical SDSs are identified and studied. Then, state-of-the-art DM models are modified to be used in scenarios which are unique to long-term personalised VUIs, such as personalised models initialised with data from different speakers or scenarios where the dialogue environment (e.g. the ASR) changes over time. In addition, several speaker and environment related features are shown to be useful to improve the interaction performance. This study is done in the context of homeService, a VUI developed to help users with dysarthria to control their home devices. The study shows that personalisation of the POMDP-DM framework can greatly improve the performance of these interfaces.

The reliability of automatically estimating human ages, by processing input facial images, has generally been found to be poor. On other hand, various real world applications, often relating to safety and security, depend on an accurate estimate of a person’s age. In such situations, Face Image based Automatic Age Estimation (FI-AAE) systems which are more reliable and may ideally surpass human ability, are of importance as and represent a critical pre-requisite technology. Unfortunately, in terms of estimation accuracy and thus performance, contemporary FI-AAE systems are impeded by challenges which exist in both of the two major FI-AAE processing phases i.e. i) Age based feature extraction and representation and ii) Age group classification. Challenges in the former phase arise because facial shape and texture change independently and the magnitude of these changes vary during the different stages of a person’s life. Additionally, contemporary schemes struggle to exploit age group specific characteristics of these features, which in turn has a detrimental effect on overall system performance. Furthermore misclassification errors which occur in the second processing phase and are caused by the smooth inter-class variations often observed between adjacent age groups, pose another major challenge and are responsible for low overall FI-AAE performance. In this thesis a novel Multi-Level Age Estimation (ML-AE) framework is proposed that addresses the aforementioned challenges and improves upon state-of-the-art FI-AAE system performance. The proposed ML-AE is a hierarchical classification scheme that maximizes and then exploits inter-class variation among different age groups at each level of the hierarchy. Furthermore, the proposed scheme exploits age based discriminating information taken from two different cues (i.e. facial shape and texture) at the decision level which improves age estimation results. During the process of achieving our main objective of age estimation, this research work also contributes to two associated image processing/analysis areas: i) Face image modeling and synthesis; a process of representing face image data with a low dimensionality set of parameters. This is considered as precursor to every face image based age estimation system and has been studied in this thesis within the context of image face recognition ii) measuring face image data variability that can help in representing/ranking different face image datasets according to their classification difficulty level. Thus a variability measure is proposed that can also be used to predict the classification performance of a given face recognition system operating upon a particular input face dataset. Experimental results based on well-known face image datasets revealed the superior performance of our proposed face analysis, synthesis and face image based age classification methodologies, as compared to that obtained from conventional schemes.

This thesis is about improving machine lip-reading, that is, the classification of speech from only visual cues of a speaker. Machine lip-reading is a niche research problem in both areas of speech processing and computer vision. Current challenges for machine lip-reading fall into two groups: the content of the video, such as the rate at which a person is speaking or; the parameters of the video recording for example, the video resolution. We begin our work with a literature review to understand the restrictions current technology limits machine lip-reading recognition and conduct an experiment into resolution affects. We show that high definition video is not needed to successfully lip-read with a computer. The term 'viseme' is used in machine lip-reading to represent a visual cue or gesture which corresponds to a subgroup of phonemes where the phonemes are indistinguishable in the visual speech signal. Whilst a viseme is yet to be formally defined, we use the common working definition: 'a viseme is a group of phonemes with identical appearance on the lips'. A phoneme is the smallest acoustic unit a human can utter. Because there are more phonemes per viseme, mapping between the units creates a many-to-one relationship. Many mappings have been presented, and we conduct an experiment to determine which mapping produces the most accurate classification. Our results show Lee's [82] is best. Lee's classification also outperforms machine lip-reading systems which use the popular Fisher [48] phoneme-to-viseme map. Further to this, we propose three methods of deriving speaker-dependent phoneme-to-viseme maps and compare our new approaches to Lee's. Our results show the sensitivity of phoneme clustering and we use our new knowledge for our first suggested augmentation to the conventional lip-reading system. Speaker independence in machine lip-reading classification is another unsolved obstacle. It has been observed, in the visual domain, that classifiers need training on the test subject to achieve the best classification. Thus machine lip-reading is highly dependent upon the speaker. Speaker independence is the opposite of this, or in other words, is the classification of a speaker not present in the classifier's training data. We investigate the dependence of phoneme-to-viseme maps between speakers. Our results show there is not a high variability of visual cues, but there is high variability in trajectory between visual cues of an individual speaker with the same ground truth. This implies a dependency upon the number of visemes within each set for each individual. Finally, we investigate how many visemes is the optimum number within a set. We show the phoneme-to-viseme maps in literature rarely have enough visemes and the optimal number, which varies by speaker, ranges from 11 to 35. The last difficulty we address is decoding from visemes back to phonemes and into words. Traditionally this is completed using a language model. The language model unit is either: the same as the classifier, e.g. visemes or phonemes; or the language model unit is words. In a novel approach we use these optimum range viseme sets within hierarchical training of phoneme labelled classifiers. This new method of classifier training demonstrates significant increase in classification with a word language network.

Over recent decades, the textured coating provided by metallic surfaces has been an important factor in attracting customers of the automobile industry. This has meant that quantifying the appearance of coating products is essential for product development and quality control. The appearance of these coated products strongly depends on the viewing geometry, giving rise to a variety of properties of perceptual attributes such as texture, colour and gloss. Due to the visually-complex nature of such coatings, there remains an unsatisfied demand to develop techniques to measure the total appearance of metallic coatings. This study describes which aims to define the total appearance of metallic coatings and then objectively characterise it. Total appearance here refers to the combination of three properties of perceptual attributes of the surface: glint, coarseness and brightness. A number of metallic panels were visually scaled and a computational model capable for predicting three perceptual attributes was developed. A computational model was developed to relate the results from this psychophysical experiment to data obtained from a stereo image capture system. This is a new alternative technique aimed at solving one of the most challenging problems in computer vision: stereo matching. In the system, two images are captured by a same camera under two different lighting conditions to mimic stereoscopic vision. This not only addresses the problem of stereo matching (i.e. to find the corresponding pixels between two images) but also enhances the effect of perceptual attributes. After linearisation of camera response, spatial uniformity correction was performed to minimise the effect of uneven illumination. A characterisation method was then used to transfer the RGB to device-independent values. Two images captured under different lighting conditions were merged to obtain stereo data. In glint feature extraction, the pixels in the final image were segmented into two regions: bright spots and dark background. Next, statistical analyses were applied to extract features. Finally a model was created to predict the glint attribute of the metallic coating panels based on an image captured by the stereo capture system. In coarseness feature extraction, the merged image transformed to frequency domain using a discrete Fourier Transform. An octave bandpass filter was then applied to the Fourier Spectra image and data analysis was carried out to achieve the “image variance value” for each band. In similar to final step of glint, a model was created to predict the coarseness attribute.

Speech is a desirable communication method between humans and computers. The major concerns of the automatic speech recognition (ASR) are determining a set of classification features and finding a suitable recognition model for these features. Hidden Markov Models (HMMs) have been demonstrated to be powerful models for representing time varying signals. Artificial Neural Networks (ANNs) have also been widely used for representing time varying quasi-stationary signals. Arabic is one of the oldest living languages and one of the oldest Semitic languages in the world, it is also the fifth most generally used language and is the mother tongue for roughly 200 million people. Arabic speech recognition has been a fertile area of reasearch over the previous two decades, as attested by the various papers that have been published on this subject. This thesis investigates phoneme and acoustic models based on Deep Neural Networks (DNN) and Deep Echo State Networks for multi-dialect Arabic Speech Recognition. Moreover, the TIMIT corpus with a wide variety of American dialects is also aimed to evaluate the proposed models. The availability of speech data that is time-aligned and labelled at phonemic level is a fundamental requirement for building speech recognition systems. A developed Arabic phoneme database (APD) was manually timed and phonetically labelled. This dataset was constructed from the King Abdul-Aziz Arabic Phonetics Database (KAPD) database for Saudi Arabia dialect and the Centre for Spoken Language Understanding (CSLU2002) database for different Arabic dialects. This dataset covers 8148 Arabic phonemes. In addition, a corpus of 120 speakers (13 hours of Arabic speech) randomly selected from the Levantine Arabic dialect database that is used for training and 24 speakers (2.4 hours) for testing are revised and transcription errors were manually corrected. The selected dataset is labelled automatically using the HTK Hidden Markov Model toolkit. TIMIT corpus is also used for phone recognition and acoustic modelling task. We used 462 speakers (3.14 hours) for training and 24 speakers (0.81 hours) for testing. For Automatic Speech Recognition (ASR), a Deep Neural Network (DNN) is used to evaluate its adoption in developing a framewise phoneme recognition and an acoustic modelling system for Arabic speech recognition. Restricted Boltzmann Machines (RBMs) DNN models have not been explored for any Arabic corpora previously. This allows us to claim priority for adopting this RBM DNN model for the Levantine Arabic acoustic models. A post-processing enhancement was also applied to the DNN acoustic model outputs in order to improve the recognition accuracy and to obtain the accuracy at a phoneme level instead of the frame level. This post process has significantly improved the recognition performance. An Echo State Network (ESN) is developed and evaluated for Arabic phoneme recognition with different learning algorithms. This investigated the use of the conventional ESN trained with supervised and forced learning algorithms. A novel combined supervised/forced supervised learning algorithm (unsupervised adaptation) was developed and tested on the proposed optimised Arabic phoneme recognition datasets. This new model is evaluated on the Levantine dataset and empirically compared with the results obtained from the baseline Deep Neural Networks (DNNs). A significant improvement on the recognition performance was achieved when the ESN model was implemented compared to the baseline RBM DNN model’s result. The results show that the ESN model has a better ability for recognizing phonemes sequences than the DNN model for a small vocabulary size dataset. The adoption of the ESNs model for acoustic modeling is seen to be more valid than the adoption of the DNNs model for acoustic modeling speech recognition, as ESNs are recurrent models and expected to support sequence models better than the RBM DNN models even with the contextual input window. The TIMIT corpus is also used to investigate deep learning for framewise phoneme classification and acoustic modelling using Deep Neural Networks (DNNs) and Echo State Networks (ESNs) to allow us to make a direct and valid comparison between the proposed systems investigated in this thesis and the published works in equivalent projects based on framewise phoneme recognition used the TIMIT corpus. Our main finding on this corpus is that ESN network outperform time-windowed RBM DNN ones. However, our developed system ESN-based shows 10% lower performance when it was compared to the other systems recently reported in the literature that used the same corpus. This due to the hardware availability and not applying speaker and noise adaption that can improve the results in this thesis as our aim is to investigate the proposed models for speech recognition and to make a direct comparison between these models.