Connections between music artists or songs provide a context and lineage for music and form the basis for recommendation, playlist generation, and general navigation of the musical universe. We examine the structure of the connections between music artists found on the web. It is shown that different methods of finding associations between artists yeild different net- work structures - the details of associations and how these associations are discovered impact the global structure of the artist network. This realization informs our associations framework - based on seman- tic web technologies and centered around a small RDF/OWL ontology that emphasizes the provenance and transparency of association statements. We develop the MuSim Similarity Ontology and show how, combined with the concepts of linked data, it can be used to create a distributed web-scale ecosystem for music similarity. The Similarity Ontology is evaluated against psychological models for similarity and shown to be flexible enough to accommodate each model examined. Several applications are developed based on the visualization of music artist network structures and the utilization of our associations framework along with other music-related linked data.

Object detection in images and action detection in videos are among the most widely studied computer vision problems, with applications in consumer photography, surveillance, and automatic media tagging. Typically, these standard detectors are fully supervised, that is they require a large body of training data where the locations of the objects/actions in images/videos have been manually annotated. With the emergence of digital media, and the rise of high-speed internet, raw images and video are available for little to no cost. However, the manual annotation of object and action locations remains tedious, slow, and expensive. As a result there has been a great interest in training detectors with weak supervision where only the presence or absence of object/action in image/video is needed, not the location. This thesis presents approaches for weakly supervised learning of object/action detectors with a focus on automatically annotating object and action locations in images/videos using only binary weak labels indicating the presence or absence of object/action in images/videos. First, a framework for weakly supervised learning of object detectors in images is presented. In the proposed approach, a variation of multiple instance learning (MIL) technique for automatically annotating object locations in weakly labelled data is presented which, unlike existing approaches, uses inter-class and intra-class cue fusion to obtain the initial annotation. The initial annotation is then used to start an iterative process in which standard object detectors are used to refine the location annotation. Finally, to ensure that the iterative training of detectors do not drift from the object of interest, a scheme for detecting model drift is also presented. Furthermore, unlike most other methods, our weakly supervised approach is evaluated on data without manual pose (object orientation) annotation. Second, an analysis of the initial annotation of objects, using inter-class and intra-class cues, is carried out. From the analysis, a new method based on negative mining (NegMine) is presented for the initial annotation of both object and action data. The NegMine based approach is a much simpler formulation using only inter-class measure and requires no complex combinatorial optimisation but can still meet or outperform existing approaches including the previously pre3 sented inter-intra class cue fusion approach. Furthermore, NegMine can be fused with existing approaches to boost their performance. Finally, the thesis will take a step back and look at the use of generic object detectors as prior knowledge in weakly supervised learning of object detectors. These generic object detectors are typically based on sampling saliency maps that indicate if a pixel belongs to the background or foreground. A new approach to generating saliency maps is presented that, unlike existing approaches, looks beyond the current image of interest and into images similar to the current image. We show that our generic object proposal method can be used by itself to annotate the weakly labelled object data with surprisingly high accuracy.

No description available.

All-optical Wave Division Multiplexed (WDM) networking is a promising technology for long-haul backbone and large metropolitan optical networks in order to meet the non-diminishing bandwidth demands of future applications and services. Examples could include archival and recovery of data to/from Storage Area Networks (i.e. for banks), High bandwidth medical imaging (for remote operations), High Definition (HD) digital broadcast and streaming over the Internet, distributed orchestrated computing, and peak-demand short-term connectivity for Access Network providers and wireless network operators for backhaul surges. One desirable feature is fast and automatic provisioning. Connection (lightpath) provisioning in optically switched networks requires both route computation and a single wavelength to be assigned for the lightpath. This is called Routing and Wavelength Assignment (RWA). RWA can be classified as static RWA and dynamic RWA. Static RWA is an NP-hard (non-polynomial time hard) optimisation task. Dynamic RWA is even more challenging as connection requests arrive dynamically, on-the-fly and have random connection holding times. Traditionally, global-optimum mathematical search schemes like integer linear programming and graph colouring are used to find an optimal solution for NP-hard problems. However such schemes become unusable for connection provisioning in a dynamic environment, due to the computational complexity and time required to undertake the search. To perform dynamic provisioning, different heuristic and stochastic techniques are used. Particle Swarm Optimisation (PSO) is a population-based global optimisation scheme that belongs to the class of evolutionary search algorithms and has successfully been used to solve many NP-hard optimisation problems in both static and dynamic environments. In this thesis, a novel PSO based scheme is proposed to solve the static RWA case, which can achieve optimal/near-optimal solution. In order to reduce the risk of premature convergence of the swarm and to avoid selecting local optima, a search scheme is proposed to solve the static RWA, based on the position of swarm‘s global best particle and personal best position of each particle. To solve dynamic RWA problem, a PSO based scheme is proposed which can provision a connection within a fraction of a second. This feature is crucial to provisioning services like bandwidth on demand connectivity. To improve the convergence speed of the swarm towards an optimal/near-optimal solution, a novel chaotic factor is introduced into the PSO algorithm, i.e. CPSO, which helps the swarm reach a relatively good solution in fewer iterations. Experimental results for PSO/CPSO based dynamic RWA algorithms show that the proposed schemes perform better compared to other evolutionary techniques like genetic algorithms, ant colony optimization. This is both in terms of quality of solution and computation time. The proposed schemes also show significant improvements in blocking probability performance compared to traditional dynamic RWA schemes like SP-FF and SP-MU algorithms.

Body-Centric Wireless Communication (BCWC) is a central point in the development of fourth generation mobile communications. The continuous miniaturisation of sensors, in addition to the advancement in wearable electronics, embedded software, digital signal processing and biomedical technologies, have led to a new concept of usercentric networks, where devices can be carried in the user’s pockets, attached to the user’s body or even implanted. Body-centric wireless networks take their place within the personal area networks, body area networks and body sensor networks which are all emerging technologies that have a broad range of applications such as healthcare and personal entertainment. The major difference between BCWC and conventional wireless systems is the radio channel over which the communication takes place. The human body is a hostile environment from radio propagation perspective and it is therefore important to understand and characterise the effect of the human body on the antenna elements, the radio channel parameters and hence the system performance. This is presented and highlighted in the thesis through a combination of experimental and electromagnetic numerical investigations, with a particular emphasis to the numerical analysis based on the finite-difference time-domain technique. The presented research work encapsulates the characteristics of the narrowband (2.4 GHz) and ultra wide-band (3-10 GHz) on-body radio channels with respect to different digital phantoms, body postures, and antenna types hence highlighting the effect of subject-specific modelling, static and dynamic environments and antenna performance on the overall body-centric network. The investigations covered extend further to include in-body communications where the radio channel for telemetry with medical implants is also analysed by considering the effect of different digital phantoms on the radio channel characteristics. The study supports the significance of developing powerful and reliable numerical modelling to be used in conjunction with measurement campaigns for a comprehensive understanding of the radio channel in body-centric wireless communication. It also emphasises the importance of considering subject-specific electromagnetic modelling to provide a reliable prediction of the network performance.

Concealed weapon detection (CWD) has been a hot topic as the concern about pub- lic safety increases. A variety of approaches for the detection of concealed objects on the human body based on earth magnetic ¯eld distortion, inductive magnetic ¯eld, acoustic and ultrasonic, electromagnetic resonance, MMW (millimetre wave), THz, Infrared, x-ray technologies have been suggested and developed. Among all of them, MMW holographic imaging is considered as a promising approach due to the relatively high penetration and high resolution that it can o®er. Typical concealed target detection methods are classi¯ed into 2 categories, the ¯rst one is a resonance based target identi¯cation technique, and the second one is an imaging based system. For the former, the complex natural resonance (CNR) frequencies associated with a certain target are extracted and used for identi¯cation, but this technique has an issue of high false alarm rate. The microwave/millimetre wave imaging systems can be categorized into two types: passive systems and active sys- tems. For the active microwave/millimetre wave imaging systems, the microwave holographic imaging approach was adopted in this thesis. Such a system can oper- ate at either a single frequency or multiple frequencies (wide band). An active, coherent, single frequency operation millimetre wave imaging system based on the theory of microwave holography was developed. Based on literature surveys and ¯rst hand experimental results, this thesis aims to provide system level parame- ter determination to aid the development of a target detection imager. The goal is approached step by step in 7 chapters, with topics and issues addressed rang- ing from reviewing the past work, ¯nding out the best candidate technology, i.e. the MMW holographic imaging combined with the resonance based target recog- i nition technique, the construction of the 94 GHz MMW holographic prototype imager, experimental trade-o® investigation of system parameters, imager per- formance evaluation, low pro¯le components and image enhancement techniques, feasibility investigation of resonance based technique, to system implementation based on the parameters and results achieved. The task set forth in the beginning is completed by coming up with an entire system design in the end. i

One of the great challenges in optics is to break the diffraction limit to achieve optical superresolution for applications in imaging, sensing, manufacturing and characterization. In recent years we witnessed a number of exciting developments in this field, including for example super-resolution fluorescent microscopy, negative-index metamaterial superlens and superoscillation lens. However, none of them can perform white-light super-resolution imaging until the development of microsphere nanoscopy technique, which was pioneered by the current PhD’s research group. The microscope nanoscopy technique was developed based on all-dielectric microsphere superlens which is fundamentally different from metal-based superlenses. In this research, we aim to significantly advance the technology by: (1) increasing superlens resolution to sub- 50 nm scale and (2) improving superlens usability and demonstrate application in wider context including lab-on-chip devices. Our longer-term vision is to bring the all-dielectric superlens technology to market so that each microscope user can have superlens in hand for their daily examination of nanoscale objects including viruses. To improve the superlens resolution, a systematic theoretical study was first carried out on the optical properties of dielectric microsphere superlens. New approaches were proposed to obtain precise control of the focusing properties of the microsphere lens. Using pupil mask engineering and two-material composite superlens design, one can precisely control the focusing properties of the lens and effectively surpass the diffraction limit λ/2n. To further improve the resolution, we incorporated the metamaterial concept in our superlens design. A new all-dielectric nanoparticle metamaterial superlens design was proposed. This is realized by 3D stacking of high-index nanoparticles to form a micro-sized particle lens. This man-made superlens has unusual optical properties not found in nature: highly effective conversion of evanescent wave to propagating wave for unprecedented optical super-resolution. By using 15 nm TiO2 nanoparticles as building blocks, the fabricated 3D all-dielectric metamaterial-based solid immersion lens (mSIL) can produce a sharp image with a super-resolution of at least 45 nm under a white-light optical microscope, significantly exceeding the classical diffraction limit and previous near-field imaging techniques. In additional to mSIL where only one kind of nanoparticle was used, we also studied twoVII nanomaterial hybrid system. High-quality microspheres consisting of ZrO2/polystyrene elements were synthesised and studied. We show precise tuning of the refractive index of microspheres can effectively enhance the imaging resolution and quality. To increase superlens usability and application scope, we proposed and demonstrated a new microscope objective lens that features a two times resolution improvement over conventional objective. This is accomplished by integrating a conventional microscope objective lens with a superlensing microsphere lens with a customised lens adaptor. The new objective lens was successfully demonstrated for label-free super-resolution static and scanning imaging of 100 nm features in engineering and biological samples. In an effort to reduce superlens technology entrance barrier, we studied several spider silks as naturally occurring optical superlens. These spider silks are transparent in nature and have micron-scale cylinder structure. They can distinctly resolve λ/6 features with a large field-of-view under a conventional white-light microscope. This discovery opens a new door to develop biology-based optical systems and has enriched the superlens category. Because microsphere superlenses are small in size, their application can be extended to lab-on-chip device. In this thesis, microsphere superlens was introduced to a microfluidic channel to build an on-chip microfluidic superlensing device for real-time high-resolution imaging of biological objects. Several biological samples with different features in size, transparency, low contrast and strong mobility have been visualised. This integrated device provides a new way to allow researchers to directly visualise details of biological specimens in real-time under a conventional white light microscope. The work carried out in this research has significantly improved the microsphere superlens technology which opens the door for commercial exploitation.

Energy is increasingly becoming an exclusive commodity in next generation wireless communication systems, where even in legacy systems, the mobile operators operational expenditure is largely attributed to the energy bill. However, as the amount of mobile traffic is expected to double over the next decade as we enter the Next Generation communications era, the need to address energy efficient protocols will be a priority. Therefore, we will need to revisit the design of the mobile network in order to adopt a proactive stance towards reducing the energy consumption of the network. Future emerging communication paradigms will evolve towards Next Generation mobile networks, that will not only consider a new air interface for high broadband connectivity, but will also integrate legacy communications (LTE/LTE-A, IEEE 802.11x, among others) networks to provide a ubiquitous communication platform, and one that can host a multitude of rich services and applications. In this context, one can say that the radio access network will predominantly be OFDMA based, providing the impetus for further research studies on how this technology can be further optimized towards energy efficiency. In fact, advanced approaches towards both energy and spectral efficient design will still dominate the research agenda. Taking a step towards this direction, LTE/LTE-A (Long Term Evolution-Advanced) have already investigated cooperative paradigms such as SON (self-Organizing Networks), Network Sharing, and CoMP (Coordinated Multipoint) transmission. Although these technologies have provided promising results, some are still in their infancy and lack an interdisciplinary design approach limiting their potential gain. In this thesis, we aim to advance these future emerging paradigms from a resource allocation perspective on two accounts. In the first scenario, we address the challenge of load balancing (LB) in OFDMA networks, that is employed to redistribute the traffic load in the network to effectively use spectral resources throughout the day. We aim to reengineer the load-balancing (LB) approach through interdisciplinary design to develop an integrated energy efficient solution based on SON and network sharing, what we refer to as SO-LB (Self-Organizing Load balancing). Obtained simulation results show that by employing SO-LB algorithm in a shared network, it is possible to achieve up to 15-20% savings in energy consumption when compared to LTE-A non-shared networks. The second approach considers CoMP transmission, that is currently used to enhance cell coverage and capacity at cell edge. Legacy approaches mainly consider fundamental scheduling policies towards assigning users for CoMP transmission. We build on these scheduling approaches towards a cross-layer design that provide enhanced resource utilization, fairness, and energy saving whilst maintaining low complexity, in particular for broadband applications.

Knowledge based methodologies

This thesis addresses the problem of integrating sensors and actuators for closed-loop control of a robotic assembly cell. In addition to the problems of interfacing the physical components of the work-cell, the difficulties of representing sensory feedback at a high level within the robot control program are investigated. A new level of robot programming, called sensor-level programming, is introduced. In this, the movements of the actuators are not given explicitly, but rather are inferred by the programming system to achieve new sensor conditions given by the programmer. Control of each sensor and actuator is distributed through a master-slave hierarchy, with each sensor and actuator having its own slave controller. A protocol for information interchange between each controller and the master is defined. If possible, the control of the kinematics of a robot arm is achieved through the manufacturer's existing control system. Under these circumstances, the actuator slave would be acting as an interface between the generic command codes issued from the central controller, and the syntax of the corresponding control instructions required by the commercial system. Sensor information is preprocessed in the sensor slaves and a set of high-level descriptors, called attributes, are sent to the central controller. Closed-loop control is achieved on the basis of these attributes. The processing of sensor information which is corrupted by noise is investigated. Sources of sensor noise are identified and new algorithms are developed to quantify the noise based on information obtained from the closed-loop servoing. Once the relative magnitudes of the system and measurement noise have been estimated, a Kalman filter is used to weight the sensor information and hence reduce the credibility given to noisy sensors; in the limit ignoring the information completely. The improvements in system performance by processing the sensor information in this way are demonstrated. The sensor-level representation and automatic error processing are embedded in a software control system, which can be used to interface commercial systems as well as purpose-built devices. An'industrial research project associated with the lay-up of carbon-fibre provides an example of its operation. A list of publications resulting from the work in this thesis is given in Appendix E.