The thesis presents a computational model for reasoning with partial information which uses default rules or information about what normally happens. The idea is to provide a means of filling the gaps in an incomplete world view with the most plausible assumptions while allowing for the retraction of conclusions should they subsequently turn out to be incorrect. The model can be used both to reason from a given knowledge base of default rules, and to aid in the construction of such knowledge bases by allowing their designer to compare the consequences of his design with his own default assumptions. The conclusions supported by the proposed model are justified by the use of a probabilistic semantics for default rules in conjunction with the application of a rational means of inference from incomplete knowledge the principle of maximum entropy (ME). The thesis develops both the theory and algorithms for the ME approach and argues that it should be considered as a general theory of default reasoning. The argument supporting the thesis has two main threads. Firstly, the ME approach is tested on the benchmark examples required of nonmonotonic behaviour, and it is found to handle them appropriately. Moreover, these patterns of commonsense reasoning emerge as consequences of the chosen semantics rather than being design features. It is argued that this makes the ME approach more objective, and its conclusions more justifiable, than other default systems. Secondly, the ME approach is compared with two existing systems: the lexicographic approach (LEX) and system Z+. It is shown that the former can be equated with ME under suitable conditions making it strictly less expressive, while the latter is too crude to perform the subtle resolution of default conflict which the ME approach allows. Finally, a program called DRS is described which implements all systems discussed in the thesis and provides a tool for testing their behaviours.

The recent popularity of surveillance video systems, specially located in urban scenarios, demands the development of visual techniques for monitoring purposes. A primary step towards intelligent surveillance video systems consists on automatic object classification, which still remains an open research problem and the keystone for the development of more specific applications. Typically, object representation is based on the inherent visual features. However, psychological studies have demonstrated that human beings can routinely categorise objects according to their behaviour. The existing gap in the understanding between the features automatically extracted by a computer, such as appearance-based features, and the concepts unconsciously perceived by human beings but unattainable for machines, or the behaviour features, is most commonly known as semantic gap. Consequently, this thesis proposes to narrow the semantic gap and bring together machine and human understanding towards object classification. Thus, a Surveillance Media Management is proposed to automatically detect and classify objects by analysing the physical properties inherent in their appearance (machine understanding) and the behaviour patterns which require a higher level of understanding (human understanding). Finally, a probabilistic multimodal fusion algorithm bridges the gap performing an automatic classification considering both machine and human understanding. The performance of the proposed Surveillance Media Management framework has been thoroughly evaluated on outdoor surveillance datasets. The experiments conducted demonstrated that the combination of machine and human understanding substantially enhanced the object classification performance. Finally, the inclusion of human reasoning and understanding provides the essential information to bridge the semantic gap towards smart surveillance video systems.

As machines become more and more portable, and part of our everyday life, it becomes apparent that developing interactive and ubiquitous systems is an important aspect of new music applications created by the research community. We are interested in developing a robust layer for the automatic annotation of audio signals, to be used in various applications, from music search engines to interactive installations, and in various contexts, from embedded devices to audio content servers. We propose adaptations of existing signal processing techniques to a real time context. Amongst these annotation techniques, we concentrate on low and mid-level tasks such as onset detection, pitch tracking, tempo extraction and note modelling. We present a framework to extract these annotations and evaluate the performances of different algorithms. The first task is to detect onsets and offsets in audio streams within short latencies. The segmentation of audio streams into temporal objects enables various manipulation and analysis of metrical structure. Evaluation of different algorithms and their adaptation to real time are described. We then tackle the problem of fundamental frequency estimation, again trying to reduce both the delay and the computational cost. Different algorithms are implemented for real time and experimented on monophonic recordings and complex signals. Spectral analysis can be used to label the temporal segments; the estimation of higher level descriptions is approached. Techniques for modelling of note objects and localisation of beats are implemented and discussed. Applications of our framework include live and interactive music installations, and more generally tools for the composers and sound engineers. Speed optimisations may bring a significant improvement to various automated tasks, such as automatic classification and recommendation systems. We describe the design of our software solution, for our research purposes and in view of its integration within other systems.

This project explores the effect of materials processing on the optical, morphological, and electrical properties of ZnO nanorods synthesised using the low-temperature (90˚C) aqueous chemical technique. A highly-alkaline (pH 11) growth solution fabricated nanorods that exhibit morphological sensitivity to the anneal atmosphere used. This was attributed to unreacted precursors trapped throughout the nanorod bulk and near the surface. A significant increase in the c-axis peak intensity post-annealing and evidence of nitrogen-doping in all annealed nanorods, confirmed the precursors were present prior to the annealing process. In addition, intense green photoluminescence was observed under UV excitation and was shown to be dependent on the anneal atmosphere. The origin of this emission was related to zinc vacancy defects that were energetically favoured during the oxygen-rich synthesis and anneal conditions according to first principle calculations. To better study the electrical properties of the ZnO nanorods they were incorporated into p-n heterostructures using p-type CuSCN. An alternative spray-coating method was developed for depositing CuSCN that was a significant improvement over previous methods as demonstrated by the high hole mobility 70 cm2/V.s. The current popularity of conductive polymers led to the comparison of hybrid inorganic-organic (ZnO-PEDOT:PSS) and purely inorganic (ZnO-CuSCN) devices. These were tested as UV photodetectors and differences in device structure were shown to have a significant impact on the device response time and responsivity. A rectification ratio of 21500 at ±3 V was achieved for these ZnO-CuSCN devices. The inorganic ZnO-CuSCN device exhibited photovoltaic behaviour at zero-bias, which highlighted it as a suitable choice for self-powered UV photodetection. The effect of processing on the photodetector performance was investigated for two sets of nanorods; pH 6 and pH 11. Consequently, a maximum photocurrent response of 30 μA (for 6 mW cm-2 irradiance) was achieved for nitrogen-annealed pH 11-nanorods with a rise time of 25 ns. The high response was assigned to fewer zinc vacancies acting as electron trap states and the introduction of N-related donor defects.

Witnessing the omnipresence of digital video media, the research community has raised the question of its meaningful use and management. Stored in immense multimedia databases, digital videos need to be retrieved and structured in an intelligent way, relying on the content and the rich semantics involved. Current Content Based Video Indexing and Retrieval systems face the problem of the semantic gap between the simplicity of the available visual features and the richness of user semantics. This work focuses on the issues of efficiency and scalability in video indexing and retrieval to facilitate a video representation model capable of semantic annotation. A highly efficient algorithm for temporal analysis and key-frame extraction is developed. It is based on the prediction information extracted directly from the compressed domain features and the robust scalable analysis in the temporal domain. Furthermore, a hierarchical quantisation of the colour features in the descriptor space is presented. Derived from the extracted set of low-level features, a video representation model that enables semantic annotation and contextual genre classification is designed. Results demonstrate the efficiency and robustness of the temporal analysis algorithm that runs in real time maintaining the high precision and recall of the detection task. Adaptive key-frame extraction and summarisation achieve a good overview of the visual content, while the colour quantisation algorithm efficiently creates hierarchical set of descriptors. Finally, the video representation model, supported by the genre classification algorithm, achieves excellent results in an automatic annotation system by linking the video clips with a limited lexicon of related keywords.

Data stored in individual autonomous databases often needs to be combined and interrelated. For example, in the Inland Water (IW) environment monitoring domain, the spatial and temporal variation of measurements of different water quality indicators stored in different databases are of interest. Data from multiple data sources is more complex to combine when there is a lack of metadata in a computation forin and when the syntax and semantics of the stored data models are heterogeneous. The main types of information retrieval (IR) requirements are query transparency and data harmonisation for data interoperability and support for multiple user views. A combined Semantic Web based and Agent based distributed system framework has been developed to support the above IR requirements. It has been implemented using the Jena ontology and JADE agent toolkits. The semantic part supports the interoperability of autonomous data sources by merging their intensional data, using a Global-As-View or GAV approach, into a global semantic model, represented in DAML+OIL and in OWL. This is used to mediate between different local database views. The agent part provides the semantic services to import, align and parse semantic metadata instances, to support data mediation and to reason about data mappings during alignment. The framework has applied to support information retrieval, interoperability and multi-lateral viewpoints for four European environmental agency databases. An extended GAV approach has been developed and applied to handle queries that can be reformulated over multiple user views of the stored data. This allows users to retrieve data in a conceptualisation that is better suited to them rather than to have to understand the entire detailed global view conceptualisation. User viewpoints are derived from the global ontology or existing viewpoints of it. This has the advantage that it reduces the number of potential conceptualisations and their associated mappings to be more computationally manageable. Whereas an ad hoc framework based upon conventional distributed programming language and a rule framework could be used to support user views and adaptation to user views, a more formal framework has the benefit in that it can support reasoning about the consistency, equivalence, containment and conflict resolution when traversing data models. A preliminary formulation of the formal model has been undertaken and is based upon extending a Datalog type algebra with hierarchical, attribute and instance value operators. These operators can be applied to support compositional mapping and consistency checking of data views. The multiple viewpoint system was implemented as a Java-based application consisting of two sub-systems, one for viewpoint adaptation and management, the other for query processing and query result adjustment.

This thesis focuses on the development of automatic accompaniment systems. We investigate previous systems and look at a range of approaches that have been attempted for the problem of beat tracking. Most beat trackers are intended for the purposes of music information retrieval where a `black box' approach is tested on a wide variety of music genres. We highlight some of the diffculties facing offline beat trackers and design a new approach for the problem of real-time drum tracking, developing a system, B-Keeper, which makes reasonable assumptions on the nature of the signal and is provided with useful prior knowledge. Having developed the system with offline studio recordings, we look to test the system with human players. Existing offline evaluation methods seem less suitable for a performance system, since we also wish to evaluate the interaction between musician and machine. Although statistical data may reveal quantifiable measurements of the system's predictions and behaviour, we also want to test how well it functions within the context of a live performance. To do so, we devise an evaluation strategy to contrast a machine-controlled accompaniment with one controlled by a human. We also present recent work on a real-time multiple pitch tracking, which is then extended to provide automatic accompaniment for harmonic instruments such as guitar. By aligning salient notes in the output from a dual pitch tracking process, we make changes to the tempo of the accompaniment in order to align it with a live stream. By demonstrating the system's ability to align offline tracks, we can show that under restricted initial conditions, the algorithm works well as an alignment tool.

The process of musical composition is sometimes conceived of as an individual, internal, cognitive process in which notation plays a passive role of transmitting or recording musical ideas. This thesis questions the role played by representations in musical composition practices. We begin by tracing how, historically, compositional practices have co-evolved with musical representations and technologies for music production. We present case studies to show that the use of graphical sketches is a characteristic feature of the early stages of musical composition and that this practice recurs across musical genres ranging from classical music to contemporary electroacoustic composition. We describe the processes involved in sketching activities within the framework of distributed cognition and distinguish an intermediate representational role for sketches that is different from what is ‘in the head’ of the composer and from the functions of more formal musical notations. Using evidences from the case studies, we argue in particular that as in other creative design processes, sketches provide strategically ambiguous, heterogeneous forms of representation that exploit vagueness, indeterminacy and inconsistency in the development of musical ideas. Building on this analysis of the functions of sketching we describe the design and implementation of a new tool, the Music Sketcher, which attempts to provide more under-specified and flexible forms of ‘sketch’ representation than are possible with contemporary composition tools. This tool is evaluated through a series of case studies which explore how the representations constructed with the tool are interpreted and what role they play in the compositional process. We show that the program provides a similar level of vagueness to pen and paper, while also facilitating re-representation and re-interpretation, thus helping bridge the gap between early representations and later stages of commitment.

In this thesis we address the subject of automatic extraction of harmony information from audio recordings. We focus on chord symbol recognition and methods for evaluating algorithms designed to perform that task. We present a novel six-dimensional model for equal tempered pitch space based on concepts from neo-Riemannian music theory. This model is employed as the basis of a harmonic change detection function which we use to improve the performance of a chord recognition algorithm. We develop a machine readable text syntax for chord symbols and present a hand labelled chord transcription collection of 180 Beatles songs annotated using this syntax. This collection has been made publicly available and is already widely used for evaluation purposes in the research community. We also introduce methods for comparing chord symbols which we subsequently use for analysing the statistics of the transcription collection. To ensure that researchers are able to use our transcriptions with confidence, we demonstrate a novel alignment algorithm based on simple audio fingerprints that allows local copies of the Beatles audio files to be accurately aligned to our transcriptions automatically. Evaluation methods for chord symbol recall and segmentation measures are discussed in detail and we use our chord comparison techniques as the basis for a novel dictionary-based chord symbol recall calculation. At the end of the thesis, we evaluate the performance of fifteen chord recognition algorithms (three of our own and twelve entrants to the 2009 MIREX chord detection evaluation) on the Beatles collection. Results are presented for several different evaluation measures using a range of evaluation parameters. The algorithms are compared with each other in terms of performance but we also pay special attention to analysing and discussing the benefits and drawbacks of the different evaluation methods that are used.

As computers become more popular in the home and workplace, sharing resources and Internet access locally is a necessity. The simplest method of choice is by deploying a Wireless Local Area Network; they are inexpensive, easy to configure and require minimal infrastructure. The wireless local area network of choice is the IEEE 802.11 standard; IEEE 802.11, however, is now being implemented on larger scales outside of the original scope of usage. The realistic usage spans from small scale home solutions to commercial ‘hot spots,’ providing access within medium size areas such as cafés, and more recently blanket coverage in metropolitan. Due to increasing Internet availability and faster network access, in both wireless and wired, the concept of using such networks for real-time services such as internet telephony is also becoming popular. IEEE 802.11 wireless access is shared with many clients on a single channel and there are three non-overlapping channels available. As more stations communicate on a single channel there is increased contention resulting in longer delays due to the backoff overhead of the IEEE 802.11 protocol and hence loss and delay variation; not desirable for time critical traffic. Simulation of such networks demands super-computing resource, particularly where there are over a dozen clients on a given. Fortunately, the author has access to the UK’s super computers and therefore a clear motivation to develop a state of the art analytical model with the required resources to validate. The goal was to develop an analytical model to deal with realistic IEEE 802.11 deployments and derive results without the need for super computers. A network analytical model is derived to model the characteristics of the IEEE 802.11 protocol from a given scenario, including the number of clients and the traffic load of each. The model is augmented from an existing published saturated case, where each client is assumed to always have traffic to transmit. The nature of the analytical model is to allow stations to have a variable load, which is achieved by modifying the existing models and then to allow stations to operate with different traffic profiles. The different traffic profiles, for each station, is achieved by using the augmented model state machine per station and distributing the probabilities to each station’s state machine accordingly. To address the gap between the analytical models medium access delay and standard network metrics which include the effects of buffering traffic, a queueing model is identified and augmented which transforms the medium access delay into standard network metrics; delay, loss and jitter. A Quality of Experience framework, for both computational and analytical results, is investigated to allow the results to be represented as user perception scores and the acceptable voice call carrying capacity found. To find the acceptable call carrying capacity, the ITU-T G.107 E-Model is employed which can be used to give each client a perception rating in terms of user satisfaction. PAGE 4 OF 162 QUEEN MARY, UNIVERSITY OF LONDON OLIVER SHEPHERD With the use of a novel framework, benchmarking results show that there is potential to maximise the number of calls carried by the network with an acceptable user perception rating. Dimensioning of the network is undertaken, again compared with simulation from the super computers, to highlight the usefulness of the analytical model and framework and provides recommendations for network configurations, particularly for the latest Wireless Multimedia extensions available in IEEE 802.11. Dimensioning shows an overall increase of acceptable capacity of 43%; from 7 to 10 bidirectional calls per Access Point by using a tuned transmission opportunity to allow each station to send 4 packets per transmission. It is found that, although the accuracy of the results from the analytical model is not precise, the model achieves a 1 in 13,000 speed up compared to simulation. Results show that the point of maximum calls comes close to simulation with the analytical model and framework and can be used as a guide to configure the network. Alternatively, for specific capacity figures, the model can be used to home-in on the optimal region for further experiments and therefore achievable with standard computational resource, i.e. desktop machines.