Mobile notifications provide a unique mechanism for real-time information delivery systems to users in order to increase its effectiveness. However, real-time notification delivery to users via mobile phones does not always translate into users' awareness about the delivered information because these alerts might arrive at inappropriate times and situations. Moreover, notifications that demand users' attention at inopportune moments are more likely to have adverse effects and become a cause of potential disruption rather than proving beneficial to users. In order to address these challenges it is of paramount importance to devise intelligent notification mechanisms that monitor and learn users' behaviour for maximising their receptivity to the delivered information and adapt accordingly. The central goal of this dissertation is to build a framework for intelligent notifications that relies on the awareness of users' context and preferences. More specifically, we firstly investigate the impact of physical and cognitive contextual features on users' attentiveness and receptivity to notifications. Secondly, we construct and evaluate a series of models for predicting opportune moments to deliver notifications and mining users' notification delivery preferences in different situations. Finally, we design and evaluate a model for anticipating the right device notifications in cross-platform environments.

A large number of approaches have been developed to simplify construction of, and to reduce errors in, data-driven applications. However, these approaches have not been particularly concerned with compile-time type safety. Type mismatch errors between program and the database schema occur quite often during program development, and the techniques used in these approaches often defer error checking on database operations until runtime. In this thesis, we take a different approach from those previously proposed, based on strict type checking at compile time, type inference, higher-order functions, phantom types, object relational mapping, and loosely coupled database interaction. Instead of using external, literal XML file and string type SQL, we embed the mapping meta data and user defined queries directly in the program, the type safety of which is guaranteed by the program compiler. Such a result is achieved by introducing additional database schema information and using type avatars, a dummy structure used to extend the type checking to embedded queries, during compilation. We show that this approach is practical and effective by implementing a compile-time type-safe object relational framework, called Qanat, in the OCaml programming language and using a loosely coupled SQL database. We further report experimental results obtained by running a number of benchmark tests, and compare the resulting Qanat applications with the equivalent, raw database driver based applications.

Every day environments contain a great variety of deformable objects and it is not possible to program a robot in advance to know about their characteristic behaviours. For this reason, robots have been highly successful in manoeuvring deformable objects mainly in the industrial sector, where the types of interactions are predictable and highly restricted, but research in everyday environments remains largely unexplored. The contributions of this thesis are: i) the application of an elastic/plastic mass-spring method to model and predict the behaviour of deformable objects manipulated by a robot; ii) the automatic calibration of the parameters of the model, using images of real objects as ground truth; iii) the use of piece-wise regression curves to predict the reaction forces, and iv) the use of the output of this force prediction model as input for the mass-spring model which in turn predicts object deformations; v) the use of the obtained models to solve a material classification problem, where the robot must recognise a material based on interaction with it.

External, tree-based, multidimensional access methods typically attempt to provide B+ tree like behaviour and performance in the organisation of large collections of multidimensional data. The B+ tree’s efficiency comes directly from the fact that it organises data occupying a single dimension, which can be linearly ordered, and partitioned at arbitrary points in that order. Using a multiway tree to partition a multidimensional space becomes increasingly difficult with increasing dimensionality, often leading to the loss of desirable properties like high fanout and low internode overlap. The K-D-B tree is an example of a structure in which one property, that of zero internode overlap, is provided at the expense of another, high fanout. Its approach to doing this, by forced splitting, is shared by a collection of other structures, and in 1995 Freeston suggested a novel approach to mitigate the effects of forced splits, by executing them virtually. This approach has not been taken up widely, but we believe it shows a great deal of promise. In the thesis, we examine the virtual forced splitting approach in depth. We identify a number of problems presented by the approach, and propose solutions to them, allowing us to characterise a general class of virtual forced splitting structures that we call VFS-trees. The efficacy of our approach is demonstrated by our implementation of a new VFS structure, and by what we believe to be the first implementation of a BV-tree, together with new algorithms for region and K Nearest Neighbour search. We further report experimental results on construction, exact-match search and K-NN search of BV-trees, and show how they compare, very favourably, with the corresponding operations on the currently most popular multidimensional file access method, the R*-tree.

Multi-agent systems operating in complex domains crucially require agents to interact with each other. An important result of this interaction is that some of the private knowledge of the agents is being shared in the group of agents. This thesis investigates the theme of knowledge sharing from a theoretical point of view by means of the formal tools provided by modal logic. More specifically this thesis addresses the following three points. First, the case of hypercube systems, a special class of interpreted systems as defined by Halpern and colleagues, is analysed in full detail. It is here proven that the logic S5WDn constitutes a sound and complete axiomatisation for hypercube systems. This logic, an extension of the modal system S5n commonly used to represent knowledge of a multi-agent system, regulates how knowledge is being shared among agents modelled by hypercube systems. The logic S5WDn is proven to be decidable. Hypercube systems are proven to be synchronous agents with perfect recall that communicate only by broadcasting, in separate work jointly with Ron van der Meyden not fully reported in this thesis. Second, it is argued that a full spectrum of degrees of knowledge sharing can be present in any multi-agent system, with no sharing and full sharing at the extremes. This theme is investigated axiomatically and a range of logics representing a particular class of knowledge sharing between two agents is presented. All the logics but two in this spectrum are proven complete by standard canonicity proofs. We conjecture that these two remaining logics are not canonical and it is an open problem whether or not they are complete. Third, following a influential position paper by Halpern and Moses, the idea of refining and checking of knowledge structures in multi-agent systems is investigated. It is shown that, Kripke models, the standard semantic tools for this analysis are not adequate and an alternative notion, Kripke trees, is put forward. An algorithm for refining and checking Kripke trees is presented and its major properties investigated. The algorithm succeeds in solving the famous muddy-children puzzle, in which agents communicate and reason about each other's knowledge. The thesis concludes by discussing the extent to which combining logics, a promising new area in pure logic, can provide a significant boost in research for epistemic and other theories for multi-agent systems.

This thesis develops a new approach to evolving Binary Decision Diagrams, and uses it to study evolvability issues. For reasons that are not yet fully understood, current approaches to artificial evolution fail to exhibit the evolvability so readily exhibited in nature. To be able to apply evolvability to artificial evolution the field must first understand and characterise it; this will then lead to systems which are much more capable than they are currently. An experimental approach is taken. Carefully crafted, controlled experiments elucidate the mechanisms and properties that facilitate evolvability, focusing on the roles and interplay between neutrality, modularity, gradualism, robustness and diversity. Evolvability is found to emerge under gradual evolution as a biased distribution of functionality within the genotype-phenotype map, which serves to direct phenotypic variation. Neutrality facilitates fitness-conserving exploration, completely alleviating local optima. Population diversity, in conjunction with neutrality, is shown to facilitate the evolution of evolvability. The search is robust, scalable, and insensitive to the absence of initial diversity. The thesis concludes that gradual evolution in a search space that is free of local optima by way of neutrality can be a viable alternative to problematic evolution on multi-modal landscapes.

Dependency Parsing is a method that builds dependency trees consisting of binary relations that describe the syntactic role of words in sentences. Recently, dependency parsing has seen large improvements due to deep learning, which enabled richer feature representations and flexible architectures. In this thesis we focus on the application of these methods to Transition-based parsing, which is a faster variant. We explore current architectures and examine ways to improve their representation capabilities and final accuracies. Our first contribution is an improvement on the basic architecture at the heart of many current parsers. We show that using Recurrent Neural Network hidden layers, initialised with pretrained weights from a feed forward network, provides significant accuracy improvements. Second, we examine the best parser architecture. We show that separate classifiers for dependency parsing and labelling, with a shared input layer provides the best accuracy. We also show that a parser and labeller can be successfully trained separately. Finally, we propose Recursive LSTM Trees, which can represent an entire tree as a single dense vector, and achieve competitive accuracy with minimal features. The parsers that we develop in this thesis cover many aspects of this task, and are easy to integrate with current methods.

This thesis studies the diversity issue of classification ensembles for class imbalance learning problems. Class imbalance learning refers to learning from imbalanced data sets, in which some classes of examples (minority) are highly under-represented comparing to other classes (majority). The very skewed class distribution degrades the learning ability of many traditional machine learning methods, especially in the recognition of examples from the minority classes, which are often deemed to be more important and interesting. Although quite a few ensemble learning approaches have been proposed to handle the problem, no in-depth research exists to explain why and when they can be helpful. Our objectives are to understand how ensemble diversity affects the classification performance for a class imbalance problem according to single-class and overall performance measures, and to make best use of diversity to improve the performance. As the first stage, we study the relationship between ensemble diversity and generalization performance for class imbalance problems. We investigate mathematical links between single-class performance and ensemble diversity. It is found that how the single-class measures change along with diversity falls into six different situations. These findings are then verified in class imbalance scenarios through empirical studies. The impact of diversity on overall performance is also investigated empirically. Strong correlations between diversity and the performance measures are found. Diversity shows a positive impact on the recognition of the minority class and benefits the overall performance of ensembles in class imbalance learning. Our results help to understand if and why ensemble diversity can help to deal with class imbalance problems. Encouraged by the positive role of diversity in class imbalance learning, we then focus on a specific ensemble learning technique, the negative correlation learning (NCL) algorithm, which considers diversity explicitly when creating ensembles and has achieved great empirical success. We propose a new learning algorithm based on the idea of NCL, named AdaBoost.NC, for classification problems. An ``ambiguity" term decomposed from the 0-1 error function is introduced into the training framework of AdaBoost. It demonstrates superiority in both effectiveness and efficiency. Its good generalization performance is explained by theoretical and empirical evidences. It can be viewed as the first NCL algorithm specializing in classification problems. Most existing ensemble methods for class imbalance problems suffer from the problems of overfitting and over-generalization. To improve this situation, we address the class imbalance issue by making use of ensemble diversity. We investigate the generalization ability of NCL algorithms, including AdaBoost.NC, to tackle two-class imbalance problems. We find that NCL methods integrated with random oversampling are effective in recognizing minority class examples without losing the overall performance, especially the AdaBoost.NC tree ensemble. This is achieved by providing smoother and less overfitting classification boundaries for the minority class. The results here show the usefulness of diversity and open up a novel way to deal with class imbalance problems. Since the two-class imbalance is not the only scenario in real-world applications, multi-class imbalance problems deserve equal attention. To understand what problems multi-class can cause and how it affects the classification performance, we study the multi-class difficulty by analyzing the multi-minority and multi-majority cases respectively. Both lead to a significant performance reduction. The multi-majority case appears to be more harmful. The results reveal possible issues that a class imbalance learning technique could have when dealing with multi-class tasks. Following this part of analysis and the promising results of AdaBoost.NC on two-class imbalance problems, we apply AdaBoost.NC to a set of multi-class imbalance domains with the aim of solving them effectively and directly. Our method shows good generalization in minority classes and balances the performance across different classes well without using any class decomposition schemes. Finally, we conclude this thesis with how the study has contributed to class imbalance learning and ensemble learning, and propose several possible directions for future research that may improve and extend this work.

Delivering environmentally friendly services has become an important issue in Cloud Computing due to awareness provided by governments and environmental conservation organisations about the impact of electricity usage on carbon footprints. Cloud providers and cloud consumers (organisations/ enterprises) have their own defined \(green\) \(policies\) to control energy consumption at their data centers. At service management level, \(green\) \(policies\) can be mapped as \(energy\) \(management\) \(policies\) or \(management\) \(policies\). Focusing at cloud consumer's side, \(management\) \(policies\) are described by business managers which can change regularly. The continuous changing is based on the nature of the technical environment, changes in regulation; and business requirements. Therefore, there is a gap between the level of describing and implementing \(management\) \(policies\) in the cloud environment. This thesis provides a method to bridge that gap by (a) defining a specification for formulating \(management\) \(policies\) into executable form for an infrastructure-as-a-service (IaaS) cloud model; (b) designing a framework to execute the described \(management\) \(policies\) automatically; (c) proposing a modelling and analysis method to identify the potential \(energy\) \(management\) \(policy\) that would save energy-cost. Each aspect covered in the thesis is evaluated with a help of an Energy Management Case Study for a private cloud scenario.

Semi-supervised classification consists of acquiring knowledge from both labelled and unlabelled data to classify test instances. The cluster assumption represents one of the potential relationships between true classes and data distribution that semi-supervised algorithms assume in order to use unlabelled data. Ensemble algorithms have been widely and successfully employed in both supervised and semi-supervised contexts. In this Thesis, we focus on the cluster assumption to study ensemble learning based on a new cluster regularisation technique for multi-class semi-supervised classification. Firstly, we introduce a multi-class cluster-based classifier, the Cluster-based Regularisation (Cluster- Reg) algorithm. ClusterReg employs a new regularisation mechanism based on posterior probabilities generated by a clustering algorithm in order to avoid generating decision boundaries that traverses high-density regions. Such a method possesses robustness to overlapping classes and to scarce labelled instances on uncertain and low-density regions, when data follows the cluster assumption. Secondly, we propose a robust multi-class boosting technique, Cluster-based Boosting (CBoost), which implements the proposed cluster regularisation for ensemble learning and uses ClusterReg as base learner. CBoost is able to overcome possible incorrect pseudo-labels and produces better generalisation than existing classifiers. And, finally, since there are often datasets with a large number of unlabelled instances, we propose the Efficient Cluster-based Boosting (ECB) for large multi-class datasets. ECB extends CBoost and has lower time and memory complexities than state-of-the-art algorithms. Such a method employs a sampling procedure to reduce the training set of base learners, an efficient clustering algorithm, and an approximation technique for nearest neighbours to avoid the computation of pairwise distance matrix. Hence, ECB enables semi-supervised classification for large-scale datasets.