In this thesis a number of hierarchical ensemble classification approaches are proposed as a solution to the multi-class classification problem. The central idea is that a more effective classification can be produced if a “coarse-grain” classification (directed at groups of classes) is first conducted followed by increasingly more “fine-grain” classifications. The Hierarchical ensemble classification model comprises a set of base classifiers held within the nodes of the hierarchy (one classifier per node). Nodes near the root hold classifiers designed to discriminate between groups of class labels while the leaves hold classifiers designed to distinguish between individual class labels. Two types of hierarchy (structures) are considered, Binary Tree (BT) hierarchies and Directed Acyclic Graph (DAG) hierarchies. With respect to the DAG structure, two alternative DAG structures to support the generation of the desired hierarchical ensemble classification model are considered: (i) rooted DAG, and (ii) non-rooted DAG. The main challenges are: (i) how best to distribute class labels between nodes within the hierarchy, (ii) how to address the “successive mis-classification” issue associated with hierarchical classification where if a mis-classication occurs early on in the process (near the root of the hierarchy) there is no possibility of rectifying this error later on in the process, and (iii) how best to determine the starting node within the non-rooted DAG approach. To address the first issue different techniques, based on the concepts of clustering, splitting, and combination, are proposed. To address the second and the third issues the idea is to utilise probability or confidence values associated with Naive Bayes and CARM classifiers respectively to dictate whether single or multiple paths should be followed at each hierarchy node, and to select the best starting DAG node with respect to the non-rooted DAG approach.

Heterogeneous processing technologies provide opportunities for new levels of flexibility and performance in applications. Current utilisation of such technologies has focused on HPC workloads and often couples applications to specific device targets through low level code. While this approach can produce optimal performance for specialised applications on a fixed hardware platform, it requires significant expertise to code and is not portable. This work presents an approach that reduces development overheads for high-level application developers and provides a framework to allow applications to dynamically target heterogeneous devices. A task based approach to application development is presented that decouples high-level code from device specific logic and also removes any need to manually manage devices from within applications. The SHEPARD framework facilitates this through a central repository of device specific implementations, and couples these with execution time estimates to effectively allocate tasks among shared heterogeneous processors. The approach presented in this work is applied to various In-Memory Database scenarios including analytic workloads. Experiments show how a single task call can enable applications to execute code on multiple devices, and managing such tasks among a limited set of devices can allow multiple user workloads to share resources effectively, while improving workload performance.

No description available.

A variety of meshless methods have been developed in the last fifteen years with an intention to solve practical engineering problems, but are limited to small academic problems due to associated high computational cost as compared to the standard finite element methods (FEM). The main objective of this thesis is the development of an efficient and accurate algorithm based on meshless methods for the solution of problems involving both material and geometrical nonlinearities, which are of practical importance in many engineering applications, including geomechanics, metal forming and biomechanics. One of the most commonly used meshless methods, the element-free Galerkin method (EFGM) is used in this research, in which maximum entropy shape functions (max-ent) are used instead of the standard moving least squares shape functions, which provides direct imposition of the essential boundary conditions. Initially, theoretical background and corresponding computer implementations of the EFGM are described for linear and nonlinear problems. The Prandtl-Reuss constitutive model is used to model elasto-plasticity, both updated and total Lagrangian formulations are used to model finite deformation and consistent or algorithmic tangent is used to allow the quadratic rate of asymptotic convergence of the global Newton-Raphson algorithm. An adaptive strategy is developed for the EFGM for two- and three-dimensional nonlinear problems based on the Chung \& Belytschko error estimation procedure, which was originally proposed for linear elastic problems. A new FE-EFGM coupling procedure based on max-ent shape functions is proposed for linear and geometrically nonlinear problems, in which there is no need of interface elements between the FE and EFG regions or any other special treatment, as required in the most previous research. The proposed coupling procedure is extended to become adaptive FE-EFGM coupling for two- and three-dimensional linear and nonlinear problems, in which the Zienkiewicz \& Zhu error estimation procedure with the superconvergent patch recovery method for strains and stresses recovery are used in the FE region of the problem domain, while the Chung \& Belytschko error estimation procedure is used in the EFG region of the problem domain. Parallel computer algorithms based on distributed memory parallel computer architecture are also developed for different numerical techniques proposed in this thesis. In the parallel program, the message passing interface library is used for inter-processor communication and open-source software packages, METIS and MUMPS are used for the automatic domain decomposition and solution of the final system of linear equations respectively. Separate numerical examples are presented for each algorithm to demonstrate its correct implementation and performance, and results are compared with the corresponding analytical or reference results.

The growth of the Internet attracted many users either as merchants or customers. Merchants have products and services to sell whereas customers have money to pay for these products and services. Customers normally do not trust merchants and also merchants do not trust customers. When a customer is willing to buy any product or service, they want be sure that the merchant will send them the product that they want after they make the correct payment. Similarly, a merchant will not take the risk by sending the product before the customer makes the correct payment. Therefore, the party (customer or merchant) who send their item (payment or product) first will be at risk of the misbehaviour of the other party. This problem is known as the fairness problem. The aim is to study the existing fair exchange protocols that solve the fairness problem. Then, propose more efficient protocols to solve the fairness problem. The original idea in this thesis is enforcing honesty in fair exchange protocols. The idea of enforcing honesty is applied to produce a fair exchange protocol that encourages the merchant to be honest and enforces the customer to be honest and vice versa. The thesis shows that it is not possible to enforce both the customer and the merchant to be honest at the same time. Hence, it proposes a third fair exchange protocol that encourages the customer and the merchant to be honest. The protocols make use of a Trusted Third Party (TIP) but its use is kept to minimum when disputes arise. In this respect they are optimistic fair exchange protocols. The proposed protocols have the following features: (1) only three messages are required to be exchanged between a customer and a merchant that apply the idea of enforcing a party to be honest; (2) the protocols guarantee strong fairness for both customer and merchant; (3) they allow both parties (customer and merchant) to check the correctness of the item of the other party before they send their item; (4) disputes are resolved automatically online by a Trusted Third Party (TIP); and (5) they are efficient in that they have a low number of modular exponentiations (which is the most expensive operation). Applying the idea of enforcing a party to be honest has helped in proposing efficient (air exchange protocols for the exchange of payments and digital products between customers and merchants.

Three readback signal detection methods are investgated for real-time flying height or headd isk spacingv ariation measuremenut nder vibration conditions. This is carried out by theoretical analysis, numerical simulation, and experimental study. The first method (amplitude detection) provides a simple way to study the head disk spacing change. The second method (PW50 parameter estimation) can be used effectively for real-time spacing variation measurement in normally operated hard disk drives, primarily in low frequency spacing variation conditions. The third method (thermal signal detection), on the other hand, is more effective and suitable for high frequency spacing variation measurement. By combining the PW50 estimation and thermal signal detection methods, a noval spacing variation detection method for the whole frequency range is constructed. This combined signal detection method not only has been used to study the head disk spacing variation itself, but also has the potential of being used for real time flying height control. Analytical models are developed for head disk assembly and head position servo control mechanisms to analyse the operation failure of hard disk drives under vibration conditions. Theoretical analysis and numerical simulation show their good agreement with experimental results. A novel active flying height control method is proposed to suppress the flying height or head-disk spacing variation in hard disk drives under vibration conditions. Simulation results show that this active flying height control can effectively suppress the head-disk spacing variation, therefore the perfon-nance and reliability of HDDs can be well improved when working in vibration conditions. * The method has a good potential to be applied to future ruggedized hard disk drives.

No description available.

Autonomous Mobile Programs (AMPs) are mobile agents that are aware of their resource needs and sensitive to the execution environment. AMPs are unusual in that, instead of using some external load management system, each AMP periodically recalculates network and program parameters and independently moves to a new location if it provides a better execution environment. Dynamic load management emerges from the behaviour of collections of AMPs. AMPs have previously been measured using mobile languages like Java Voyager on local area networks (LANs). The thesis develops an accurate simulation for AMPs on networks and validates it by reproducing the behaviour of collections of AMPs on homogeneous and heterogeneous LANs. The analysis shows that AMPs exhibit thrashing like other distributed load balancers. This thrashing is investigated in collections of AMPs, and two types of redundant movement (greedy effect) are identified. The thesis explores the extent of greedy effects by simulating collections of AMPs, and proposes negotiating AMPs (NAMPs) to ameliorate the problem. The design of AMPs with a competitive negotiation scheme (cNAMPs) is presented, followed by a performance comparison AMPs and cNAMPs using simulation. To estimate the significance of the greedy effects the properties of balanced states are established, such as independent balance, singleton optimality, and consecutive optimality. The balanced states are characterised for homogeneous and heterogeneous networks where AMPs are analysed as the general case. The significance of the cNAMP greedy effect is established by conducting a worst case analysis of redundant movements, and the maximum number, and probability of, redundant movements are calculated for homogeneous and heterogeneous networks. One of three theorems proves that in a heterogeneous network of q subnetworks the number of redundant movements does not exceed q − 1. i The thesis proposes and evaluates a multilevel cNAMP architecture that abstracts over network topologies to effectively distribute cNAMPs in large networks. The thesis investigates alternatives for implementation of this multilevel architecture and proposes a fusion-based scheme where information is first available to neighbour nodes. These neighbour nodes modify the information and pass it to remote locations. The effectiveness of the scheme is evaluated by simulating networks with up to five levels, varying the number of locations from 5 to 336, and the number of cNAMPs from 8 to 3360. The experiments investigate the effects depending on the number of levels, topologies, number of locations, number of cNAMPs, work of cNAMPs, type of cNAMPs, speed of locations, and type of rebalancing. The architecture is found to be effective because it delivers performance close to the hypothetical, e.g. each additional level increases mean cNAMP completion time by just 2%.

Bayesian analysis of full waveform laser detection and ranging (LaDAR) signals using reversible jump Markov chain Monte Carlo (RJMCMC) algorithms have shown higher estimation accuracy, resolution and sensitivity to detect weak signatures for 3D surface profiling, and construct multiple layer images with varying number of surface returns. However, it is computational expensive. Although parallel computing has the potential to reduce both the processing time and the requirement for persistent memory storage, parallelizing the serial sampling procedure in RJMCMC is a significant challenge in both statistical and computing domains. While several strategies have been developed for Markov chain Monte Carlo (MCMC) parallelization, these are usually restricted to fixed dimensional parameter estimates, and not obviously applicable to RJMCMC for varying dimensional signal analysis. In the statistical domain, we propose an effective, concurrent RJMCMC algorithm, state space decomposition RJMCMC (SSD-RJMCMC), which divides the entire state space into groups and assign to each an independent RJMCMC chain with restricted variation of model dimensions. It intrinsically has a parallel structure, a form of model-level parallelization. Applying the convergence diagnostic, we can adaptively assess the convergence of the Markov chain on-the-fly and so dynamically terminate the chain generation. Evaluations on both synthetic and real data demonstrate that the concurrent chains have shorter convergence length and hence improved sampling efficiency. Parallel exploration of the candidate models, in conjunction with an error detection and correction scheme, improves the reliability of surface detection. By adaptively generating a complimentary MCMC sequence for the determined model, it enhances the accuracy for surface profiling. In the computing domain, we develop a data parallel SSD-RJMCMC (DP SSD-RJMCMCU) to achieve efficient parallel implementation on a distributed computer cluster. Adding data-level parallelization on top of the model-level parallelization, it formalizes a task queue and introduces an automatic scheduler for dynamic task allocation. These two strategies successfully diminish the load imbalance that occurred in SSD-RJMCMC. Thanks to the coarse granularity, the processors communicate at a very low frequency. The MPIbased implementation on a Beowulf cluster demonstrates that compared with RJMCMC, DP SSD-RJMCMCU has further reduced problem size and computation complexity. Therefore, it can achieve a super linear speedup if the number of data segments and processors are chosen wisely.

No description available.