Information resources abound on the Internet, but mining these resources is a non-trivial task. Such abundance has raised the need to enhance services provided to users, such as recommendations. The purpose of this work is to explore how better recommendations can be provided to specialists in specific domains such as bioinformatics by introducing semantic techniques that reason through different resources and using specialist search techniques. Such techniques exploit semantic relations and hidden associations that occur as a result of the information overlapping among various concepts in multiple bioinformatics resources such as ontologies, websites and corpora. Thus, this work introduces a new method that reasons over different bioinformatics resources and then discovers and exploits different relations and information that may not exist in the original resources. Such relations may be discovered as a consequence of the information overlapping, such as the sibling and semantic similarity relations, to enhance the accuracy of the recommendations provided on bioinformatics content (e.g. articles). In addition, this research introduces a set of semantic rules that are able to extract different semantic information and relations inferred among various bioinformatics resources. This project introduces these semantic-based methods as part of a recommendation service within a content-based system. Moreover, it uses specialists' interests to enhance the provided recommendations by employing a method that is collecting user data implicitly. Then, it represents the data as adaptive ontological user profiles for each user based on his/her preferences, which contributes to more accurate recommendations provided to each specialist in the field of bioinformatics.

by Wu Kam-wah. / Bibliography: leaves [129]-[130] / Thesis (M.Ph.)--Chinese University of Hong Kong, 1985

Communication networks are witnessing a dramatic growth in the number of connected mobile devices, sensors and the Internet of Everything (IoE) equipment, which have been estimated to exceed 50 billion by 2020, generating zettabytes of traffic each year. In addition, networks are stressed to serve the increased capabilities of the mobile devices (e.g., HD cameras) and to fulfil the users' desire for always-on, multimedia-oriented, and low-latency connectivity. To cope with these challenges, service providers are exploiting softwarised, cost-effective, and flexible service provisioning, known as Network Function Virtualisation (NFV). At the same time, future networks are aiming to push services to the edge of the network, to close physical proximity from the users, which has the potential to reduce end-to-end latency, while increasing the flexibility and agility of allocating resources. However, the heavy footprint of today's NFV platforms and their lack of dynamic, latency-optimal orchestration prevents them from being used at the edge of the network. In this thesis, the opportunities of bringing NFV to the network edge are identified. As a concrete solution, the thesis presents Glasgow Network Functions (GNF), a container-based NFV framework that allocates and dynamically orchestrates lightweight virtual network functions (vNFs) at the edge of the network, providing low-latency network services (e.g., security functions or content caches) to users. The thesis presents a powerful formalisation for the latency-optimal placement of edge vNFs and provides an exact solution using Integer Linear Programming, along with a placement scheduler that relies on Optimal Stopping Theory to efficiently re-calculate the placement following roaming users and temporal changes in latency characteristics. The results of this work demonstrate that GNF's real-world vNF examples can be created and hosted on a variety of hosting devices, including VMs from public clouds and low-cost edge devices typically found at the customer's premises. The results also show that GNF can carefully manage the placement of vNFs to provide low-latency guarantees, while minimising the number of vNF migrations required by the operators to keep the placement latency-optimal.

This thesis presents a novel and generic visual architecture for scene understanding by robotic interactive perception. This proposed visual architecture is fully integrated into autonomous systems performing object perception and manipulation tasks. The proposed visual architecture uses interaction with the scene, in order to improve scene understanding substantially over non-interactive models. Specifically, this thesis presents two experimental validations of an autonomous system interacting with the scene: Firstly, an autonomous gaze control model is investigated, where the vision sensor directs its gaze to satisfy a scene exploration task. Secondly, autonomous interactive perception is investigated, where objects in the scene are repositioned by robotic manipulation. The proposed visual architecture for scene understanding involving perception and manipulation tasks has four components: 1) A reliable vision system, 2) Camera-hand eye calibration to integrate the vision system into an autonomous robot’s kinematic frame chain, 3) A visual model performing perception tasks and providing required knowledge for interaction with scene, and finally, 4) A manipulation model which, using knowledge received from the perception model, chooses an appropriate action (from a set of simple actions) to satisfy a manipulation task. This thesis presents contributions for each of the aforementioned components. Firstly, a portable active binocular robot vision architecture that integrates a number of visual behaviours are presented. This active vision architecture has the ability to verge, localise, recognise and simultaneously identify multiple target object instances. The portability and functional accuracy of the proposed vision architecture is demonstrated by carrying out both qualitative and comparative analyses using different robot hardware configurations, feature extraction techniques and scene perspectives. Secondly, a camera and hand-eye calibration methodology for integrating an active binocular robot head within a dual-arm robot are described. For this purpose, the forward kinematic model of the active robot head is derived and the methodology for calibrating and integrating the robot head is described in detail. A rigid calibration methodology has been implemented to provide a closed-form hand-to-eye calibration chain and this has been extended with a mechanism to allow the camera external parameters to be updated dynamically for optimal 3D reconstruction to meet the requirements for robotic tasks such as grasping and manipulating rigid and deformable objects. It is shown from experimental results that the robot head achieves an overall accuracy of fewer than 0.3 millimetres while recovering the 3D structure of a scene. In addition, a comparative study between current RGB-D cameras and our active stereo head within two dual-arm robotic test-beds is reported that demonstrates the accuracy and portability of our proposed methodology. Thirdly, this thesis proposes a visual perception model for the task of category-wise objects sorting, based on Gaussian Process (GP) classification that is capable of recognising objects categories from point cloud data. In this approach, Fast Point Feature Histogram (FPFH) features are extracted from point clouds to describe the local 3D shape of objects and a Bag-of-Words coding method is used to obtain an object-level vocabulary representation. Multi-class Gaussian Process classification is employed to provide a probability estimate of the identity of the object and serves the key role of modelling perception confidence in the interactive perception cycle. The interaction stage is responsible for invoking the appropriate action skills as required to confirm the identity of an observed object with high confidence as a result of executing multiple perception-action cycles. The recognition accuracy of the proposed perception model has been validated based on simulation input data using both Support Vector Machine (SVM) and GP based multi-class classifiers. Results obtained during this investigation demonstrate that by using a GP-based classifier, it is possible to obtain true positive classification rates of up to 80\%. Experimental validation of the above semi-autonomous object sorting system shows that the proposed GP based interactive sorting approach outperforms random sorting by up to 30\% when applied to scenes comprising configurations of household objects. Finally, a fully autonomous visual architecture is presented that has been developed to accommodate manipulation skills for an autonomous system to interact with the scene by object manipulation. This proposed visual architecture is mainly made of two stages: 1) A perception stage, that is a modified version of the aforementioned visual interaction model, 2) An interaction stage, that performs a set of ad-hoc actions relying on the information received from the perception stage. More specifically, the interaction stage simply reasons over the information (class label and associated probabilistic confidence score) received from perception stage to choose one of the following two actions: 1) An object class has been identified with high confidence, so remove from the scene and place it in the designated basket/bin for that particular class. 2) An object class has been identified with less probabilistic confidence, since from observation and inspired from the human behaviour of inspecting doubtful objects, an action is chosen to further investigate that object in order to confirm the object’s identity by capturing more images from different views in isolation. The perception stage then processes these views, hence multiple perception-action/interaction cycles take place. From an application perspective, the task of autonomous category based objects sorting is performed and the experimental design for the task is described in detail.

There are well-established approaches of text mining collections of documents and for understanding the network of citations between academic papers. Few studies have examined the textual content of the papers that constitute a citation network. A document corpus was obtained from the arXiv repository, selected from papers relating to the subject of Dark Matter and a citation network was created from the data held by NASA’s Astrophysics Data System on those papers, their citations and references. I use the Louvain community-finding algorithm on the Dark Matter network to identify groups of papers with a higher density of citations and compare the textual similarity between papers in the Dark Matter corpus using the Vector Space Model of document representation and the cosine similarity function. It was found that pairs of papers within a citation community have a higher similarity than they do with papers in other citation communities. This implies that content is associated with structure in scientific citation networks, which opens avenues for research on network communities for finding ground-truth using advanced Text Mining techniques, such as Topic Modelling. It was found that using the titles of papers in a citation network community was a good method for identifying the community. The power law exponent of the degree distribution was found to be, = 2.3, lower than results reported for other citation networks. The selection of papers based on a single subject, rather than based on a journal or category, is suggested as the reason for this lower value. It was also found that the degree pair correlation of the citation network classifies it as a disassortative network with a cut-off value at degree kc = 30.The textual similarity of documents decreases linearly with age over a 15 year timespan.

Typescript (photocopy). / Digitized by Kansas Correctional Industries

Digitized by Kansas Correctional Industries

Probabilistic model checking is a quantitative verification technique that aims to verify the correctness of probabilistic systems. Nevertheless, it suffers from the so-called state space explosion problem. In this thesis, we propose two new model reduction techniques to improve the efficiency and scalability of verifying probabilistic systems, focusing on discrete-time Markov chains (DTMCs). In particular, our emphasis is on verifying quantitative properties that bound the time or cost of an execution. We also focus on methods that avoid the explicit construction of the full state space. We first present a finite-horizon variant of probabilistic bisimulation for DTMCs, which preserves a bounded fragment of PCTL. We also propose another model reduction technique that reduces what we call linear inductive DTMCs, a class of models whose state space grows linearly with respect to a parameter. All the techniques presented in this thesis were developed in the PRISM model checker. We demonstrate the effectiveness of our work by applying it to a selection of existing benchmark probabilistic models, showing that both of our two new approaches can provide significant reductions in model size and in some cases outperform the existing implementations of probabilistic verification in PRISM.

The purpose of this thesis is to apply automatic detection techniques on forward scatter radar for ground targets detection against vegetation clutter background and thermal noise. This thesis presents the FSR automatic detection performance analysis of three signal processing algorithms: coherent, non-coherent and cross-correlation. The concept of a CFAR forward scatter radar detection is presented and includes pre-fixed threshold detection and adaptive threshold detection. The developments of a set of simulation methods for target detection and performance analysis are described in details. In the results, we will compare the probability of detection for both human and vehicle target against a variety of clutter backgrounds - WGN, stationary narrow band clutter, non-stationary narrow band clutter, and real recorded vegetation clutter at a low (VHF and UHF) frequency bands. Finally, the advantages and limitations of detection performance for each signal processing algorithms are described.

The Cloud is an environment designed for the provision of on-demand resource sharing and data access to remotely located clients and devices. Once data is outsourced to the Cloud, clients tend to lose control of their data thus becoming susceptible to data theft. To mitigate/ reduce the chances of data theft, Cloud service providers employ methods like encrypting data prior to outsourcing it to the Cloud. Although this increases security, it also gives rise to the challenge of searching and sifting through the large amounts of encrypted documents present in the Cloud. This thesis proposes a comprehensive framework that provides Searchable Encryption-as-a-Service (SEaaS) by enabling clients to search for keyword(s) over the encrypted data stored in the Cloud. Searchable Encryption (SE) is a methodology based on recognized cryptographic primitives to enable a client to search over the encrypted Cloud data. This research makes five major contributions to the field of Searchable Encryption: The first contribution is that the thesis proposes novel index-based SE schemes that increase the query effectiveness while being lightweight. To increase query effectiveness this thesis presents schemes that facilitate single-keyword, parallelized disjunctive-keyword (multi-keyword) and fuzzy-keyword searches. The second contribution of this research is the incorporation of probabilistic trapdoors in all the proposed schemes. Probabilistic trapdoors enable the client to hide the search pattern even when the same keyword is searched repeatedly. Hence, this quality allows the client to resist distinguishability attacks and prevents attackers from inferring the search pattern. The third contribution is the enumeration of a "Privacy-preserving" SE scheme by presenting new definitions for SE; i.e., keyword-trapdoor indistinguishability and trapdoor index indistinguishability. The existing security definitions proposed for SE did not take into account the incorporation of probabilistic trapdoors hence they were not readily applicable to our proposed schemes; hence new definitions have been studied. The fourth contribution is the validation that the proposed index-based SE schemes are efficient and can be deployed on to the real-world Cloud offering. The proposed schemes have been implemented and proof-of-concept prototypes have been deployed onto the British Telecommunication's Cloud Server (BTCS). Once deployed onto the BTCS the proof-of-concept prototypes have been tested over a large real-world speech corpus. The fifth contribution of the thesis is the study of a novel homomorphic SE scheme based on probabilistic trapdoors for the provision of higher level of security and privacy. The proposed scheme is constructed on a Partially Homomorphic Encryption Scheme that is lightweight when compared to existing Fully Homomorphic-based SE schemes. The scheme also provides non-repudiation of the transmitted trapdoor while eliminating the need for a centralized data structure, thereby facilitating scalability across Cross-Cloud platforms.