Abstract:Freshwaters are closely linked with adjacent terrestrial ecosystems through reciprocal resource subsidies, which are fluxes of nutrients, organisms, and materials between ecosystems. Terrestrial ecosystems provide many resource subsidies to freshwaters including leaf litter, one of the most prevalent terrestrial-derived subsidies. Inputs of leaf litter fuel detritivores food web, as food resources and refuges, and affect nutrients cycling in freshwaters. The decomposition of leaf litter is subjected to many biotic and abiotic factors, which makes it a good indicator of freshwater ecosystem functioning. Yet, this ecosystem process has been affected by anthropogenic disturbances that alter abiotic and biotic factors in the nature. Therefore, this thesis aimed to investigate some previously under-investigated or unclear but important factors that may affect the decomposition of leaf litter in streams. First, I reviewed the importance of resource subsidy fluxes between riparian zones and freshwaters and how these subsidies can influence recipient ecosystems. Then, I conducted a field experiment exploring the effects of anthropogenic carrion subsidy (chicken meat) and environmental-relevant concentration of glyphosate (the most widely applied herbicides worldwide) on leaf litter decomposition and invertebrate communities colonizing in the leaf-litter bags deploying in streams with different types of land use. Next, I conducted a mesocosm experiment nearby an urban stream to investigate the effects of water temperature (~ 8 oC above vs ambient), consumer - snails (presence vs absence), and leaf-litter quality (intact vs >40 % leaf area was consumed by terrestrial insects) on litter decomposition. Finally, I explored the global patterns of riparian leaf litter C, N, P, and their stoichiometric ratios to gradients of climatic (mean annual temperature and precipitation) and geographic (absolute latitude and altitude) factors, and the differences between biotic factors (phylogeny, leaf habit, N-fixing function, invasion status, and life form). The results of field experiment indicated that: in coarse mesh bags, glyphosate, carrion subsidy, and the addition of both decreased litter breakdown rates by 6.3 %, 22.6 %, and 24.3 % respectively; in fine mesh bags, glyphosate and the addition of both retarded litter breakdown rates by 8.3 % and 12.5 % respectively. Litter decomposition also differed among streams, with the highest breakdown rates in village streams and lowest in urban/suburban streams. Invertebrates were significantly different among streams, with biodiversity index and total taxon richness were highest in village streams and lowest in suburban stream. However, overall effects of carrion subsidy and glyphosate on macroinvertebrates were not significant. The results of mesocosm experiment indicated that warming and the presence of snails accelerated litter decomposition by 60.2 % and 34.9 % respectively, while litter breakdown rates of terrestrial insect damaged leaves were 5.1 % slower than intact leaves because of lower leaf litter quality. The results of meta-analysis study demonstrated that global riparian leaf litter had higher N and P, while lower C, C:N, and C:P ratios than terrestrial leaf litter in general. Riparian leaf litter quality changed with gradients of climatic and geographic predictors, and these patterns differed between leaf habits (evergreen or deciduous) and climate zones (tropical or non-tropical area). In general, my research provides important information on resource subsidy processes, which will benefit freshwater ecosystem management to support biodiversity and maintain ecosystem services.

Abstract:Mobility is an element that is highly related to the development of society and the quality of individual life. Through mass of automobile production and traffic infrastructure construction, advanced countries have reached a high degree of individual mobility. In order to increase the efficiency, convenience and safety of mobility, advanced traffic infrastructure construction, transportation systems and automobiles should be developed. Among all the systems for modern automobiles, cameras based assistance systems are one of the most important components. Recently, with the development of driver assistance systems and autonomous cars, detection and recognition of traffic scene objects based on computer vision become more and more indispensable. On the other hand, the deep learning methods, in particular convolutional neural networks have achieved excellent performance in a variety of computer vision tasks. This thesis mainly presents the contributions to the computer vision and deep learning methods for traffic scene objects detection and recognition. The first approach develops numbers of methods for traffic sign detection and recognition. For traffic sign detection, template matching is applied with new features extended from chain code. Moreover, the region based convolutional neural networks are applied for detecting traffic signs painted on road surface. For traffic sign recognition, convolutional neural networks with a variety of architectures are trained with different training algorithms. The second approach focuses on the detection related to traffic text. A novel license plate detection framework is developed that is able to improve detection performance by simultane- ously completing detection and segmentation. Due to the larger number and complex layout of Chinese characters, Chinese traffic text detection faces more challenges than English text detection. Therefore, Chinese traffic texts are detected by applying convolutional neural networks and directed acyclic graph. The final approach develops a method for pedestrian attribute classification. Generally, there are irrelevant elements included in features of convolutional neural networks. In order to improve classification performance, a novel feature selection algorithm is developed to refine features of convolutional neural networks.

Abstract:Rapid urbanisation brings severe challenges on sustainable development and living quality of urban residents. Smart cities develop holistic solutions in the field of urban ecosystems using collected data from different types of Internet of Things (IoT) sources. Today, smart city research and applications have significantly surged as consequences of IoT and machine learning technological enhancement. As advanced machine learning methods, deep learning techniques provide an effective framework which facilitates data mining and knowledge discovery tasks especially in the area of computer vision and natural language processing. In recent years, researchers from various research fields attempted to apply deep learning technologies into smart city applications in order to establish a new smart city era. Much of the research effort on smart city has been made, for example, intelligent transportation, smart healthcare, public safety, etc. Meanwhile, we still face a lot of challenges as the deep learning techniques are still premature for smart city. In this thesis, we first provide a review of the latest research on the convergence of deep learning and smart city for data processing. The review is conducted from two perspectives: while the technique-oriented view presents the popular and extended deep learning models, the application-oriented view focuses on the representative application domains in smart cities. We then focus on two areas, which are intelligence transportation and social media analysis, to demonstrate how deep learning could be used in real-world applications by addressing some prominent issues, e.g., external knowledge integration, multi-modal knowledge fusion, semi-supervised or unsupervised learning, etc. In intelligent transportation area, an attention-based recurrent neural network is proposed to learn from traffic flow readings and external factors for multi-step prediction. More specifically, the attention mechanism is used to model the dynamic temporal dependencies of traffic flow data and a general fusion component is designed to incorporate the external factors. For the traffic event detection task, a multi-modal Generative Adversarial Network (mmGAN) is designed. The proposed model contains a sensor encoder and a social encoder to learn from both traffic flow sensor data and social media data. Meanwhile, the mmGAN model is extended to a semi-supervised architecture by leveraging generative adversarial training to further learn from unlabelled data. In social media analysis area, three deep neural models are proposed for crisis-related data classification and COVID-19 tweet analysis. We designed an adversarial training method to generate adversarial examples for image and textual social data to improve the robustness of multi-modal learning. As most social media data related to crisis or COVID-19 is not labelled, we then proposed two unsupervised text classification models on the basis of the state-of-the-art BERT model. We used the adversarial domain adaptation technique and the zero-shot learning framework to extract knowledge from a large amount of unlabeled social media data. To demonstrate the effectiveness of our proposed solutions for smart city applications, we have collected a large amount of real-time publicly available traffic sensor data from the California department of transportation and social media data (i.e., traffic, crisis and COVID-19) from Twitter, and built a few datasets for examining prediction or classification performances. The proposed methods successfully addressed the limitations of existing approaches and outperformed the popular baseline methods on these real-world datasets. We hope the work would move the relevant research one step further in creating truly intelligence for smart cities.

Abstract:Solid oxide fuel cell (SOFC) offers clean, renewable power generation with high efficiency, but it is susceptible to chromium poisoning that leads to considerable degradation of electrochemical performance. Volatile chromium species are released from Cr-containing interconnect materials, diffused, and deposited on electrodes in new phases by the interaction with electrode materials. The mechanisms of Cr poisoning are not clear completely yet, it requires studying how to alleviate Cr poisoning in SOFC.  This thesis presents several possible mechanisms and corresponding degradation phenomena on cathodes. While there have been several studies on chromium deposition on conventional La0.6Sr0.4Co0.2Fe0.8O3 (LSCF) cathode, to date little research has focused on understanding the microstructure, especially in porosity, the effect on the chromium deposition process. We analysed the microstructure of the initial ceramic at four porosities by various physisorption methods. We found that macropores dominate at porosity 50% (LSCF-50), while mesopores dominate at porosity 20% (LSCF-20). Porosity also changes the initial surface composition: Co-rich at LSCF-50 and Sr-rich at LSCF-20.  Upon Cr exposure, the phase and chemical state changes were identified by XRD, Raman spectroscopy, ICP- OES, and XPS, with respect to different porosities and ageing times. Among the Cr deposits, it appeared a novel phase, correlated with Cr substitution into LSCF lattice. The Cr deposition had three valence states of Cr, as a result of the atomic interactions and interfusion between Cr source and LSCF.  At the porous ceramic, Sr on the surface is correlated with the formation of SrCrO4, whereas the dense ceramic showed the lower concentration of SrCrO4 and favourably formation of Cr substitution into the LSCF lattice. The Cr adsorption also causes the redistribution of other cations at the surface and in bulk.  At LSCF-50, La, Fe and Co cations preferably dissolved into the bulk with ageing time, meanwhile, CoOx was formed and segregated at specific sites, associated with macropore distribution. The Cr penetration could be detected at depth up to 17 mm by EDX.  Depth profile showed the Cr concentration non-linearly decreased with depth. The Cr adsorption increased the concentration of Fe and Co in the near-surface region; moreover, the Sr enrichment was at the near-surface area and bulk. At LSCF-20, the surface concentration of La, Fe, and Co fluctuated with ageing time. Chromium deposition on LSCF at porosities at 20 and 50 had distinct kinetic mechanisms that may be caused by different gas transport preferences. In terms of porosity, the possible mechanism was proposed in the thesis. The density functional theory plus U (DFT+U) calculation method investigated the electronic structure and stability of LSCF and Cr-substituted perovskites at different spin states. Moreover, theoretical calculation predicted the decomposition products of LSCF and Cr-substituted perovskites and may explain the cation diffusion at the surface. LSCF and Cr doped LSF was suggested to decompose into bi/trinary oxides or/and simper perovskites. The decomposition pathway of Cr doped LSF greatly depends on temperature and environment. Computational approaches suggest the potential oxide electrolyte material LaSiAl5/6Ge1/6O5.083, LaSi5/6AlP1/6O5.083 and LaSi7/6Al5/6O5.083. The simulated interstitial positions suggest the possible pathway and predict the flexibility.  These insights can guide further composition optimization. The relatively high energy above the convex hull for other doping schemes of Ca2Al2SiO7, BaSi2O5 and K2Ba7Si16O40, indicates high phase instability, which suggests those are not good candidates for electrolyte material. 

Abstract:Although Virtual Reality (VR) has been developed for a while, the last decade has seen a surge in its popularity with the advent of commercial VR Head-Mounted Displays (HMDs), making the technology more accessible. One field that significantly benefits from VR is the entertainment industry, for example, games. Games can be challenging to design as they involve several components that are found in other types of applications as well, such as presentation, navigation, interaction with virtual agents, and in-game measurements. Despite recent advances, the optimal configurations for game applications in VR are still widely unexplored. In this thesis, we propose to fill this gap by a series of studies that analyse different components involved in making VR applications more enjoyable. We propose studying three characteristics that are heavily influential in game enjoyment (1) the aesthetical realism and emotions of virtual agents; (2) viewing perspective (First-Person Perspective and Third-Person Perspective), its influence on subjective feelings and how to measure those feelings; and (3) how to reduce or eliminate VR Sickness without affecting the experience (or affecting it positively). Our results showed that Virtual Agents' facial expressions are one of the most important aspects to be considered. On the second topic, we have observed that viewing perspective is influential on VR Sickness; however, other subjective feelings were challenging to measure in this context. On the last topic, we analysed existing tendencies in Simulator Sickness mitigation techniques that do not affect in-game mechanics and present a novel solution that has a good trade-off between mitigating VR Sickness and maintaining or enhancing immersion and performance. Finally, we propose some guidelines based on our results.

Abstract:Increasing evidences suggest that post-transcriptional RNA modifications regulate essential biomolecular functions and are related to the pathogenesis of various diseases. To date, the study of epitranscriptome layer gene regulation is mostly focused on the function of mediator proteins of RNA methylation limited by laborious experimental procedures. However, there is limited investigation of the functional relevance of individual m6A RNA methylation sites. To address this, we annotated human m6A sites in large-scale based on the guilt-by-association principle from complex biological networks. In the first chapter, the network was constructed based on public human MeRIP-Seq datasets profiling the m6A epitranscriptome under independent experimental conditions. By systematically examining the network characteristics obtained from the RNA methylation profiles, a total of 339,158 putative gene ontology functions associated with 1446 human m6A sites were identified. These are biological functions that may be regulated at epitranscriptome layer via reversible m6A RNA methylation. The results were further validated on a soft benchmark by comparing to a random predictor. In the second chapter, another approach was applied to annotate the individual human m6A sites by integrating the methylation profile, gene expression profile and protein-protein interaction network with guilt-by-association principle. The consensus signals on sites were amplified by multiplying the co-methylation network and the methylation-expression network. The PPI network smoothed the correlation for a query site to gene expression for furthering GSEA functional annotation. In the third chapter, we functionally annotated 18,886 m6A sites that are conserved between human and mouse from a larger epitranscriptome datasets using method previously described. Besides, we also completed two side projects related to SARS-CoV-2 viral m6A site prediction and m6A site prediction from Nanopore sequencing technology.

Abstract:A main theme in sociolinguistics is register variation, a situation and use dependent variation of language. Numerous studies have provided evidence of linguistic variation across situations of use in English. However, very little attention has been paid to the language of corporate blogs (CBs), which is often seen as an emerging genre of computer-mediated communication (CMC). Previous studies on blogs and corporate blogs have provided important information about their linguistic features as well as functions; however, our understanding of the linguistic variation in corporate blogs remains limited in particular ways, because many of these previous studies have focused on individual linguistic features, rather than how features interact and what the possible relations between forms (linguistic features) and functions are. Given these limitations, it would be necessary to have a more systematic perspective on linguistic variation in corporate blogs. In order to study register variation in corporate blogs more systematically, a combined framework rooted in Systemic Functional Linguistics (SFL), and register theories (e.g., Biber, 1988, 1995; Halliday & Hasan, 1989) is adopted. This combination is based on some common grounds they share, which concern the functional view of language, co-occurrence patterns of linguistic features, and the importance of large corpora to linguistic research. Guided by this framework, this thesis aims to: 1) investigate the functional linguistic variations in corporate blogs, and identify the text types that are distinguished linguistically, as well as how the CB text types cut across CB industry-categories, and 2) to identify salient linguistic differences across text types in corporate blogs in the configuration of the three components of the context of situation - field, tenor, and mode of discourse. In order to achieve these goals, a 590,520-word corpus consisting of 1,020 textual posts from 41 top-ranked corporate blogs is created and mapped onto the combined framework which consists of Biber’s multi-dimensional (MD) approach and Halliday’s SFL. Accordingly, two sets of empirical analyses are conducted one after another in this research project. At first, by using a corpus-based MD approach which applies multivariate statistical techniques (including factor analysis and cluster analysis) to the investigation of register variation, CB text types are identified; and then, some linguistic features, including the most common verbs and their process types, personal pronouns, modals, lexical density, and grammatical complexity, are selected from language metafunctions of mode, tenor and field within the SFL framework, and their linguistic differences across different text types are analysed. The results of these analyses not only show that the corporate blog is a hybrid genre, representing a combination of various text types, which serve to achieve different communicative purposes and functional goals, but also exhibit a close relationship between certain text types and particular industries, which means the CB texts categorized into a certain text type are mainly from a particular industry. On this basis, the lexical and grammatical features (i.e., the most common verbs, pronouns, modal verbs, lexical density and grammatical complexity) associated with Halliday’s metafunctions are further explored and compared across six text types. It is found that language features which are related to field, tenor and mode in corporate blogs demonstrate a dynamic nature: centring on an interpersonal function, the online blogs in a business setting are basically used for the purposes of sales, customer relationship management and branding. This research project contributes to the existing field of knowledge in the following ways: Firstly, it develops the methodology used in corpus investigation of language variation, and paves the way for further research into corporate blogs and other forms of electronic communication and, more generally, for researchers engaging in corpus-based investigations of other language varieties. Secondly, it adds greatly to a description of corporate blog as a language variety in its own right, which includes different text types identified in CB discourse, and some linguistic features realized in the context of situation. This highlights the fact that corporate blogs cannot be regarded as a simple discourse; rather, they vary according to text types and context of situation.

Abstract:With the continuous progress in science and technology, a myriad of implantable medical devices (IMDs) have been invented aimed at improving public health and wellbeing. One of the main problems with these devices is their limited battery lifetime. This results in otherwise unnecessary surgeries to replace depleted batteries leading to excessive medical expenses. Wireless power transfer (WPT), as a promising technology, could be used to remedy this. Wireless power technologies, both through the transfer of transmitted radio frequency (RF) power or the harvesting of RF energy from the ambient environment and its subsequent conversion to useable electrical energy, are emerging as important features for the future of electronic devices in general and have attracted an upsurge in research interest. Unfortunately, the path to realising this wire free charging dream is paved with many thorns and there still exist critical challenges to be addressed. This thesis aims to deal with some of these challenges, developing an efficient WPT system for IMDs. The work begins with a comprehensive study of currently applied methods of WPT, which broadly fall into two categories: far-field (radiative) WPT and near field (non-radiative) WPT. The review includes a brief history of WPT, comparisons between current methodologies applied and a comprehensive literature review. Magnetic resonance coupling (MRC) WPT is emphasised due to its advantages for the desired application making it the technology of choice for system development. Design of an MRC-WPT system requires an understanding of the performance of the four basic topologies available for the MRC method. Following an investigation of these, it is found that series primary circuits are generally most suitable for WPT and that the choice of a series or parallel secondary circuit is dependent on the relative size of the load impedance. Importantly, design parameters must be optimised to avoid the phenomena of frequency splitting to simultaneously obtain maximum power transfer efficiency (PTE) and load power. The use of printed spiral coils (PSCs) as inductors in the construction of WPT circuits for IMDs, which can save space and be integrated with other circuit boards, is then investigated. The challenges and issues of PSCs present for WPT mainly relate to maintaining an inductive characteristic at frequencies in the Medical Implant Communication Service (MICS) band and to maximising the PTE between primary and secondary circuits. Investigations of PSC design parameters are performed to obtain inductive characteristics at high frequencies and the split-ring loop is proposed to increase the Quality factor relative to that offered by the PSC, which is shown to enhance WPT performance. To simplify the necessary resonating circuit configuration for MRC-WPT, a self-resonating split-ring loop with a series inductor-capacitor characteristic has been developed. A pair of these self-resonators has been adopted into a series primary-series secondary WPT system operating at high frequency. This is different to traditional planar self-resonators, which offer parallel self-resonance characteristics that are less desirable due to their reduced system power insertion as a parallel primary resonator. Finally, a system for implantable devices is developed using the split-ring loop in consideration of the effects of body tissues, whose dielectric characteristics have a significant influence on WPT performance. Due concern is also paid to human safety from radiated RF power. A series resonating split-ring loop for transmitting power is formed at the desired frequency through the addition of a lumped element capacitor. A single loop as a receiving resonator with a low Specific Absorption Rate (SAR), is designed to allow greater transmit power to be used in comparison to previous work, whilst satisfying the relevant standards relating to human safety. A rectifier circuit is also designed to convert the received RF energy into useable electrical energy allowing the realisation of the proposed WPT system. In a nutshell, this thesis places emphasis on solutions to overcome challenges relating to the use of MRC-WPT for IMDs. An efficient near-field WPT system for such devices is successfully demonstrated and should have profound significance in pushing forward the future development of this topic.

Abstract:Recently, organo-lead-halide perovskite solar cells have attracted growing and wide attention due to their remarkable photoelectric properties, low cost and ease of fabrication. However, the development of perovskite solar cells is still limited by several factors, such as strict fabrication conditions, low stability, small active area and poor reproducibility etc. The nature of perovskite film formation is argued as a process of a series of chemical reactions and crystallization processes, where the Pb(II) coordination chemistry involves in. We thus set out to improve the performance of perovskite films from the point of Pb(II) coordination chemistry.   By using the solvent engineering strategy, a series of inverted perovskite solar cells (PSCs) with a device structure as ITO/PEDOT: PSS/CH3NH3PbI3-xClx/PCBM/Al via one-step coating were fabricated had been successfully fabricated using simple one-step method from the solutions of chloric precursors in the mixtures of N, N-dimethylformamide (DMF) and -butyrolactone (GBL) at different ratios. The highest average PCE (power conversion efficiency) of 11.251 % was achieved when the solvent with DMF : GBL = 3.5 : 6.5 (v : v) was used for precursor preparation, while the average PCEs for the devices from precursors with pure GBL and DMF as the solvent were 8.600 % and 8.082 %, respectively. The detailed SEM (scanning electron microscope), XRD (X-ray Diffraction) and UV-Vis (UV-Visible spectroscopy) studies showed that the great increase of the PCE of the PSC was led by the apparent quality improvement of the perovskite film, owing to the fast nucleation and the slow crystal growth introduced by the dual solvent system. Plausible formation mechanisms of perovskite films from different solvents were proposed.   The film formation processes from different precursors were also studied, and several intermediates in the perovskite film formation processes were isolated and structurally characterized. The single crystals were successfully grew and the crystal structures of MAPbI3·DMF, MAPbI2Cl·DMF and MAPb1.5I3Br·DMF were solved. The crystal structures of MAPbI2Cl·DMF and MAPb1.5I3Br·DMF were identified for the first time. Meanwhile, the recrystallization process of MAPbI2Cl·DMF were founded that happened before spin-coating or at the early stage of annealing was the key to produce the perovskite film with high crystallinity and high orientation from chloride precursors. Based on the structures and chemical properties of the intermediates, the version of chemical reactions and mechanisms for perovskite film formation with different precursors were proposed.   In addition, several groups of PSCs from lead acetate trihydrate-based precursors were constructed by varying hydrate number, finely tuning spin-coating method, and applying DMSO as additive. It was found that the H2O molecules in precursors can greatly improve the film coverage, and the pre-heating method can avoid the low crystallinity while ensuring the high coverage of perovskite thin films. In addition, the adding of DMSO as additive influenced the formation kinetics of perovskite films and improved the reproducibility of devices. As a result, PSCs with PCE of 15.714 % had been achieved.

Abstract:Supply chains are operating in increasingly volatile business environments. Supply chain resilience development has become the core task of supply chain risk management for companies to maintain effectiveness and efficiency. To achieve this, research has predominantly focused on looking for approaches to improve companies’ ability to resist, respond to, and recover from influences of disruptive events. Fundamentally, developing resilience for risk management considers three phases, including pre-, during, and post-disruption phases. Specifically, supply chain resilience research primarily investigates supply chain readiness, responsiveness, and recovery. Among the main streams of supply chain resilience research, studies predominantly focus on the conceptual development of various capabilities for each phase. Business practices only serve to elucidate the definition of supply chain resilience and capabilities. Thus, the extant research area lacks sufficient empirical understanding of how resilience is achieved in a volatile environment in industrial sectors. To address the gaps in the literature, three papers have been developed, including one literature review paper (conceptual paper) and two empirical papers. Companies from the manufacturing industry, especially the machinery sector, were selected for investigation in this research. This study investigates different aspects of supply chain resilience for efficient risk management in uncertain and volatile environments, such as international sourcing strategies to respond to a global pandemic and the role of government interventions in achieving collaborative relationships during a pandemic. Therefore, this research makes a significant theoretical contribution to the supply chain resilience and risk management literature, providing insightful, practical implications for supply chains operating in the turbulent business environment.

Abstract:The thesis presents multimodal conceptual frameworks and their applications in improving the robustness and the performance of big data analytics through cross-modal interaction or integration. A joint interpretation of several knowledge renderings such as stream, batch, linguistics, visuals and metadata creates a unified view that can provide a more accurate and holistic approach to data analytics compared to a single standalone knowledge base. Novel approaches in the thesis involve integrating multimodal framework with state-of-the-art computational models for big data, cloud computing, natural language processing, image processing, video processing, and contextual metadata. The integration of these disparate fields has the potential to improve computational tools and techniques dramatically. Thus, the contributions place multimodality at the forefront of big data analytics; the research aims at mapping and under- standing multimodal correspondence between different modalities. The primary contribution of the thesis is the Multimodal Analytics Framework (MAF), a collaborative ensemble framework for stream and batch processing along with cues from multiple input modalities like language, visuals and metadata to combine benefits from both low-latency and high-throughput. The framework is a five-step process: Data ingestion. As a first step towards Big Data analytics, a high velocity, fault-tolerant streaming data acquisition pipeline is proposed through a distributed big data setup, followed by mining and searching patterns in it while data is still in transit. The data ingestion methods are demonstrated using Hadoop ecosystem tools like Kafka and Flume as sample implementations. Decision making on the ingested data to use the best-fit tools and methods. In Big Data Analytics, the primary challenges often remain in processing heterogeneous data pools with a one-method-fits all approach. The research introduces a decision-making system to select the best-fit solutions for the incoming data stream. This is the second step towards building a data processing pipeline presented in the thesis. The decision-making system introduces a Fuzzy Graph-based method to provide real-time and offline decision-making. Lifelong incremental machine learning. In the third step, the thesis describes a Lifelong Learning model at the processing layer of the analytical pipeline, following the data acquisition and decision making at step two for downstream processing. Lifelong learning iteratively increments the training model using a proposed Multi-agent Lambda Architecture (MALA), a collaborative ensemble architecture between the stream and batch data. As part of the proposed MAF, MALA is one of the primary contributions of the research.The work introduces a general-purpose and comprehensive approach in hybrid learning of batch and stream processing to achieve lifelong learning objectives. Improving machine learning results through ensemble learning. As an extension of the Lifelong Learning model, the thesis proposes a boosting based Ensemble method as the fourth step of the framework, improving lifelong learning results by reducing the learning error in each iteration of a streaming window. The strategy is to incrementally boost the learning accuracy on each iterating mini-batch, enabling the model to accumulate knowledge faster. The base learners adapt more quickly in smaller intervals of a sliding window, improving the machine learning accuracy rate by countering the concept drift. Cross-modal integration between text, image, video and metadata for more comprehensive data coverage than a text-only dataset. The final contribution of this thesis is a new multimodal method where three different modalities: text, visuals (image and video) and metadata, are intertwined along with real-time and batch data for more comprehensive input data coverage than text-only data. The model is validated through a detailed case study on the contemporary and relevant topic of the COVID-19 pandemic. While the remainder of the thesis deals with text-only input, the COVID-19 dataset analyzes both textual and visual information in integration.  

Abstract:Increasing evidence has suggested that RNA modifications regulate many important biological processes. To date, more than 170 types of post-transcriptional RNA modifications have been discovered. With recent advances in sequencing techniques, tens of thousands of modification sites are identified in a typical high-throughput experiment, posing a key challenge to distinguish the functional modified sites from the remaining ‘passenger’ (or ‘silent’) sites. To ensure that the massive epitranscriptome datasets are properly taken advantage of, annotated, and shared, bioinformatics solutions are developed with various focuses. In this thesis, we first described a comparative conservation analysis of the human and mouse m6A epitranscriptome at single-site resolution. A novel scoring framework, ConsRM, was devised to quantitatively measure the degree of conservation of individual m6A sites. ConsRM integrates multiple information sources and a positive-unlabeled learning framework, which integrated genomic and sequence features to trace subtle hints of epitranscriptome layer conservation. With a series of validation experiments in mouse, fly and zebrafish, we showed that ConsRM outperformed well-adopted conservation scores (phastCons and phyloP) in distinguishing the conserved and non-conserved m6A sites. Additionally, the m6A sites with a higher ConsRM score are more likely to be functionally important. To further unveil the functional epitranscriptome, we investigated the potential influence of genetic factors on epitranscriptome disturbance. Recent studies have found close associations between RNA modifications and multiple pathophysiological disorders, the precise identification and large-scale prediction of disease-related modification sites can truly contribute to understanding potential disease mechanisms. Consequently, we developed a computational pipeline to systemically identify RNA modification-associated variants and their affected modification regions, with emphasis on their disease- and trait-associations. Furthermore, we described the next research considering the dynamics of RNA methylome across different tissues by elucidating the tissue-specific impact of the somatic variant on m6A methylation. The TCGA cancer mutations (derived from 27 cancer types) that may lead to the gain or loss of m6A sites in corresponding cancer-originating tissues were systemically evaluated and collected. Token together, the proposed bioinformatics pipelines and databases should serve as useful resources for functional discrimination and annotation of the massive epitranscriptome data, with implications for the potential disease mechanisms functioning through epitranscriptome layer.

Abstract:The overall objective of this work is to investigate the effect of industrial waste and low-cost material as catalysts in fast pyrolysis products upgrading and to obtain valuable chemicals. Red mud, a by-product of the Bayer process in the aluminium industry, was catalysed with beechwood for the in-situ upgrading of fast pyrolysis vapour products. It was revealed that the catalysis of beechwood with thermally pre-treated red mud enhanced the vapour upgrading effect. Individual oxides (α-Al2O3, Fe2O3, SiO2, and TiO2), the main constituents of red mud were also tested for the identification of their individual impact on the upgrading process. A biomass/catalyst weight ratio (wt. ratio) of 1:4, on the basis of relative peak area, showed the strongest effect on the product distribution. Red mud was found to reduce phenolic compounds and promote the formation of cellulose- and hemicellulose-derived furfurals and hemicellulose-derived acetic acid, which can be used for the production of a broad range of chemicals. α-Al2O3 and Fe2O3 reduced the relative yield of phenols as well, whereas the formation of furfurals was promoted by Fe2O3 and TiO2. SiO2 showed a negligible effect on fast pyrolysis vapours. The impact of catalysts on the product distribution was discussed for phenols, furfurals, and acids, for which the strongest effects were observed. This work also investigated the activity of CaO as a catalyst in the aliphatic and cyclic ketonisation reaction and depletion of phenolic compounds in the catalytic fast pyrolysis of OW. Three basic aspects were investigated: The heterogeneous character of CaO in different wt. ratios for catalytic fast pyrolysis of OW, the stability of the catalyst by re-utilisation in successive runs, and the role of H2O and CO2 in the deterioration of the catalytic performance by contact with atmospheric air. CaO catalyst promoted the selectivity for ketonisation reactions with acetone and cyclic ketones formation, whereas most of phenolic compounds were declined. The characterisation by X-ray diffraction (XRD) and Fourier transform-infrared (FT-IR) spectroscopies led to the conclusion that CaO chemisorbs significant amount of H2O and CO2 by contact with room air. It was demonstrated that CO2 was the main deactivating agent, whereas the negative effect of water was less important. The catalyst reused several runs without significant deactivation. The activation by outgassing at temperatures 950 oC was required to revert the CO2 poisoning. In order to investigate the impact of crystal structures in fast pyrolysis products upgrading, five single-phase compounds (CaTiO3, CaSiO3, Ca2Fe2O5, Ca2FeAlO5 and CaAl2O4) were synthesised and employed for catalytic upgrading of biomass fast pyrolysis vapours. All compounds did not show strong catalytic activity on the transformation of undesirable compounds into valuable compounds. However, their impact were seen in decreasing the overall yield of pyrolysis products. Finally, two types of composites (CaTiO3/CaO and Ca2Fe2O5/CaO) in different mol % were synthesised to check synergy and to prevent sintering over multiple carbonations and decarbonisation cycles on CaO catalyst and results were compared with the catalytic capability of pure CaO. It was found that synergy of CaTiO3 with CaO did not impact the catalytic performance of CaO. Besides, CaTiO3/CaO composites were found to further asssist the CaO catalytic activity in the selectivity of ketonisation reactions for acetone and cyclic ketones formation. On the contrary, the selectivity for ketonisation reactions for acetone formation decreased with incomplete conversion of acetic acid in Ca2Fe2O5/CaO composites. Furfural transformation and phenols depletion were also impacted over the presence of Ca2Fe2O5 with CaO.

Abstract:This work presents the study of a three-dimensional (3D) simulation of the concrete behaviour in a uni-axial compressive test and flexural test using discrete element modelling (DEM). The proposed numerical models are namely, unreinforced cylindrical concrete under a uni-axial compressive test, unreinforced concrete beam under three-point flexural test and lastly, steel reinforced concrete beam under four-point flexural test. Those models were built up with fish programming language and python programming language (see Appendix A1 for the code created) and run into a computer program namely Particle flow code (PFC 3D). The main aim of this paper is to validate those numerical models developed and to study the cracking initiation and failure process in order to understand the fracture behaviour of concrete. The particles were distributed using an algorithm that is based on the sieve test analysis. The parameters were set up in order to validate the numerical model with the experimental result. It was observed that all the three models developed show a strong correlation with the laboratory experiment in term of stress-strain response, load-displacement response, crack pattern and macroscopic cracks development. Once, the bond between the spheres is broken, it leads to the formation of microscopic cracks which is not visible in laboratory experiment. DEM can help to identify which part is more prone to the evolution of microscopic cracks to macroscopic cracks under the discrete fracture network. In addition to, the rosette plot allows identifying the orientation that leads to a significant amount of micro cracks which is essential for designing structures. From the observation recorded in this research, it was observed that DEM is capable to reproduce concrete behaviour both quantitatively and qualitatively. It is also possible to measure the strain energy stored in the linear contact bond and parallel bond. At yield point which corresponds to the maximum amount of microcracks recorded, that strain energy is released in the form of kinetic energy, frictional slip energy, energy of dashpot, local damping. This can be extended further to compute fracture energy in the future work. Hence, it can be concluded DEM can be used to study the heterogeneous nature of concrete and as well as randomness nature of the fracturing of concrete structure.

Abstract:Tropilaelaps mercedesae (small mite) is one of two major honey bee ectoparasitic mite species responsible for the colony losses of Apis mellifera in Asia. Although T. mercedesae mites are still restricted in Asia (except Japan), they may diffuse all over the world due to the ever-increasing global trade of live honey bees (ex. Varroa destructor). Understanding the ecological characteristics of T. mercedesae at molecular level could potentially result in improving the management and control programs. However, molecular and genomic characterization of T. mercedesae remains poorly studied, and even no genes have been deposited in Genbank to date. Therefore, I conducted T. mercedesae genome and transcriptome sequencing. By comparing T. mercedesae genome with other arthropods, I have gained new insights into evolution of Parasitiformes and the evolutionary changes associated with specific habitats and life history of honey bee ectoparasitic mite that could potentially improve the control programs of T. mercedesae. Finally, characterization of T. mercedesae transient receptor potential channel, subfamily A, member 1 (TmTRPA1) would also help us to develop a novel control method for T. mercedesae.

Abstract:In the 2008 financial crisis, many investors endured heavy financial losses caused by sharply increased volatility. The growing demand of hedging volatility via trading volatility derivatives contributes to the rapid development of VIX futures market in recent years. In this high-volatility market, initial margin is the first defense for exchanges to fight against potential default losses. Adequate initial margin can cover most losses caused by unexpected price movements, while the liquidity of market may be squeezed in the presence of high margins. Therefore, a trade-off need be balanced for exchanges in setting appropriate margins. The primary aim of this thesis is to develop an option-based framework of margin setting, standards, and evaluation in VIX futures market. I further conduct a series of comprehensive empirical studies on stochastic behavior and term structure, and margin adequacy of VIX futures. This research consists of three separate studies with distinct focuses in each. In the first study, I investigate jumps in VIX futures market, as jumps usually result in extreme stochastic behavior in VIX futures prices, which is the primary concern of both investors and regulators. I provide the empirical evidence of jumps in this market by employing three non-parametric methods for statistical testing, and further propose a nonparametric framework which is rooted in the generalized moment method (GMM) and the extreme value theory (EVT) in order to investigate the properties of jumps in VIX futures prices. As such, the magnitudes of risk premiums are quantified at price and variation levels by incorporating both VIX options and futures, showing that relatively higher premiums are paid by investors against large upward jumps in VIX futures prices, commentated with smaller premiums gaining from the variations of VIX futures. In the second study, I propose multi-factor models for VIX options and futures to study their performance in pricing VIX options and margin management on VIX futures, equipped with a set of hump-shaped volatility functions. A procedure of the generalized moment method (GMM) is developed to estimate these models by incorporating both the “forward-looking” information from VIX options and the “backward looking” information from VIX futures. The empirical results suggest that the three-factor model outperforms other candidate models in pricing VIX options, well characterize the stochastic behavior and capture the dynamics of the term structure of VIX futures. Moreover, the option-incorporated VaR and ES risk estimates can Granger-cause initial margins imposed by the Chicago Futures Exchange (CFE). In the third study, I develop an option-based framework to study margin setting, standards, and evaluation for VIX futures. More specially, the payoffs involving the trading (long/short) positions in VIX futures are converted into the ones to barrier options with moderate assumptions. By virtue of this idea, the adapted framework transforms the tier initial margin requirements to the prices of corresponding barrier options for long and short positions. I hence propose two standards of margin setting (the zero-NPV standard and zero-default-loss one) to examine adequacy of initial margins of VIX futures. These standards eventually deliver the bounds on initial margins of VIX futures, which are estimated in the risk-neutral measure. The empirical results suggest that the tier margins imposed by the CFE are sufficient to reject positive net present value (NPV) of futures positions but still not high enough to cover any possible default losses caused by the fluctuations in VIX futures prices. Furthermore, I explore the appropriate margin bounds for VIX futures with various maturities. The proposed lower and upper bounds respectively indicate the minimum and maximum required margin standards for VIX futures. The margin bounds are evaluated from the perspectives of prudentiality and opportunity cost in the physical measure, showing that the more attentions are paid to capital burden on investors in the VIX futures market, despite its volatile nature.

Abstract:In the background of China’s economic development mode being focused the worldwide attention, there is a growing trend to study the risk transmission pattern and the crisis forecasting mechanism for China’s financial markets by domestic and global academics. The study progress, however, is observed to be affected by two gaping research problems: 1) few studies construct comparative contagion models and integrated crisis forecasting systems for China’s financial markets and 2) current econometric models hired to the risk spreading effects detection and the financial crisis forecasts are yet deterministically investigated in terms of the effectiveness on China. To fill the gaps, this research proposes two hybrid contagion models and prototypes the early warning systems with motivations of first analyzing the crisis linkages and transmission channels across domestic markets in hierarchical frameworks, and then predicting the market turbulence by integrating the crisis identifying techniques and time-dependent deep learning neuron networks. To accomplish our aims, the full project is progressed in phases by solving four technical challenges that portray two literature gaps of A) the crisis identification on the basis of price volatility state distinction, B) the decomposition for multivariate correlated patterns to infer the interdependence structure and risk spillover dynamics respectively, C) the real-time warning signals generation in comparison of between traditional and stylized predictive models and D) the contagion information fusion in the EWS frameworks to distinguish the leading indicators from between internal macroeconomic factors and external risk transmitters in statistical validation metrics. The research mainly contributes to the comparative analysis on financial contagion effects detection and market turbulence prediction through the hybrid model innovations for CM and EWS development, and meanwhile brings practical significance to improve the risk management in investing activities and support the crisis prevention in policy-making. In addition, the model experimented results corroborate the China-characterized mode on risk transmissions and crisis warnings that 1) the stocks and real estate markets are verified to play the central role among risk transmitters, while the managed floating foreign exchange rate and the non-fully liberalized bond market are peripheral during the crisis; and 2) the all-round opening up policy increases the possibility of domestic security markets being exposed to external risk factors, especially relating to the cash flows, energy commodities and precious metals.

Abstract:Recently, in computer vision, a branch of machine learning, called deep learning, has attracted high attention due to its superior performance in various computer vision tasks such as image classification, object detection, semantic segmentation, action recognition and image description generation. Deep learning aims at discovering multiple levels of distributed representations, which have been validated to be discriminatively powerful in many tasks. Visual attention is an ability of the vision system to selectively focus on the salient and relevant features in a visual scene. The core objective of visual attention is to achieve the least possible amount of visual information to be processed to solve the complex high-level tasks, e.g., object recognition, which can lead the whole vision process to become effective. The visual attention is not a new topic which has been addressed in the conventional computer vision algorithms for many years. The development and deployment of visual attention in deep learning algorithms are of vital importance since the visual attention mechanism matches well with the human visual system and also shows an improving effect in many real-world applications. This thesis is on the visual attention in deep learning, starting from the recent progress in visual attention mechanism, followed by several contributions on the visual attention mechanism targeting at diverse applications in computer vision, which include the action recognition from still images, action recognition from videos and image description generation. Firstly, the soft attention mechanism, which was initially proposed to combine with Recurrent Neural Networks (RNNs), especially the Long Short-term Memories (LSTMs), ii Visual Attention Mechanism in Deep Learning and Its Applications Shiyang Yan was applied in image description generation. In this thesis, instead, as one contribution to the visual attention mechanism, the soft attention mechanism is proposed to directly plug into the convolutional neural networks for the task of action recognition from still images. Specifically, a multi-branch attention network is proposed to capture the object that the human is intereating with and the scene in which the action is performing. The soft attention mechanism applying in this task plays a significant role in capturing multi-type contextual information during recognition. Also, the proposed model can be applied in two experimental settings: with and without the bounding box of the person. The experimental results show that the proposed networks achieved state-of-the-art performance on several benchmark datasets. For the action recognition from videos, our contribution is twofold: firstly, the hard at- tention mechanism, which selects a single part of features during recognition, is essentially a discrete unit in a neural network. This hard attention mechanism shows superior capacity in discriminating the critical information/features for the task of action recognition from videos, but is often with high variance during training, as it employs the REINFORCE algorithm as its gradient estimator. Hence, this brought another critical research question, i.e., the gradient estimation of the discrete unit in a neural network. In this thesis, a Gumbel-softmax gradient estimator is applied to achieve this goal, with much lower vari- ance and more stable training. Secondly, to learn a hierarchical and multi-scale structure for the multi-layer RNN model, we embed discrete gates to control the information be- tween each layer of the RNNs. To make the model differentiable, instead of using the REINFORCE-like algorithm, we propose to use Gumbel-sigmoid to estimate the gradient of these discrete gates. For the task of image captioning, there are two main contributions in this thesis: pri- marily, the visual attention mechanism can not only be used to reason on the global image features but also plays a vital role in the selection of relevant features from the fine-grained objects appear in the image. To form a more comprehensive image representation, as a iii Visual Attention Mechanism in Deep Learning and Its Applications Shiyang Yan contribution to the encoder network for image captioning, a new hierarchical attention network is proposed to fuse the global image and local object features through the con- struction of a hierarchical attention structure, to better the visual representation for the image captioning. Secondly, to solve an inherent problem called exposure-biased issue of the RNN-based language decoder commonly used in image captioning, instead of only relying on the supervised training scheme, an adversarial training-based policy gradient op- timisation algorithm is proposed to train the networks for image captioning, with improved results on the evaluation metrics. In conclusion, comprehensive research has been carried out for the visual attention mechanism in deep learning and its applications, which include action recognition and im- age description generation. Related research topics have also been discussed, for example, the gradient estimation of the discrete units and the solution to the exposure-biased issue in the RNN-based language decoder. For the action recognition and image captioning, this thesis presents several contributions which proved to be effective in improving existing methods.

Abstract: Many researchers have argued that the welfare effect of fiscal decentralisation (FD) theory is not clearly determined outside of democratic countries. To provide more empirical evidence for the welfare effect of FD, this study focuses on the effect of China’s FD on social welfare in the post-1994 fiscal structure. Because the 1994 tax reform effectively resulted in fiscal recentralisation on the revenue side and fiscal decentralisation on the expenditure side, I construct indicators of revenue decentralisation and expenditure decentralisation and include both sets of indicators to ascertain the marginal effect of each under China’s post-1994 fiscal structure. This study consists of three chapters. In the first chapter, I examine the impact of the post-1994 fiscal structure on provincial environmental spending from 1994 to 2017. The findings show that expenditure decentralisation caused a reduction in provincial governments’ environmental spending. Although revenue decentralisation led to an increase in provincial environmental spending, its effect on the latter was not as significant as the effect of expenditure decentralisation. On balance, it can be inferred that, given China’s official promotion system and other institutional and policy factors, its post-1994 fiscal structure has had a negative effect on provincial environmental spending. In the second chapter, I explore the impact of fiscal structure on the urban-rural income gap in China after 1994, using data from 2007-2018. The findings suggest that the post-1994 fiscal structure significantly reduced the urban-rural income gap. This result is reinforced when the influence of FD on the rural-urban income gap is examined via its impact on public investment in education, healthcare and social security. The third chapter focuses on how China’s post-1994 fiscal structure has affected housing affordability in China, based on panel data from 1999 to 2017. Due to the spectacular rise of housing prices in major cities, housing has become increasingly unaffordable in China. This chapter examines the role of the fiscal structure, by which most city governments rely heavily on revenues from land-lease to help finance their many fiscal responsibilities as well as expenditure on pro-growth projects important to city officials’ promotion under China’s GDP-focused promotion system. The findings show that China’s post-1994 fiscal structure has impeded the effectiveness of its affordable housing policies. This is because the fiscal structure, together with China’s promotion system, has been an important factor behind the sharp rise in housing prices, on the one hand, and the lack of investment in affordable housing programmes, on the other. Considering that the current level of welfares can be heavily determined by those past level, the dynamic effects in these processes have been modelled by including the lagged welfare-related dependent variables on the right-hand side of the regression equations. A system generalised method of moments (Sys-GMM) estimator has been used in this research to solve the autocorrelation problem caused by the presence of lagged dependent variables while considering the endogeneity of the FD variable. In summary, the findings suggest that China’s post-1994 fiscal structure had a mixed effect on social welfare. It had a negative effect on environmental spending and housing affordability, but a positive effect on reducing urban-rural income inequality. The conclusions from this research have implications for the current debate on China’s fiscal framework and the design of future fiscal reforms. The results suggest that Chinese policymakers should consider redistributing fiscal responsibilities to match the revenue and expenditure responsibilities of all levels of government. Furthermore, the results suggest that future fiscal reforms should not only involve fiscal policy. Given that China’s official promotion system is influencing the effectiveness of sub-national officials’ decision-making, policymakers should consider adjusting the assessment criteria of the official promotion system to incentivise local officials to provide higher levels of social welfare.