The n-tuple network: coping with sub-optimality

The subject of this thesis is the n-tuple net.work (RAMnet). The major advantage of RAMnets is their speed and the simplicity with which they can be implemented in parallel hardware. On the other hand, this method is not a universal approximator and the training procedure does not involve the minimisation of a cost function. Hence RAMnets are potentially sub-optimal. It is important to understand the source of this sub-optimality and to develop the analytical tools that allow us to quantify the generalisation cost of using this model for any given data. We view RAMnets as classifiers and function approximators and try to determine how critical their lack of' universality and optimality is. In order to understand better the inherent. restrictions of the model, we review RAMnets showing their relationship to a number of well established general models such as: Associative Memories, Kamerva's Sparse Distributed Memory, Radial Basis Functions, General Regression Networks and Bayesian Classifiers. We then benchmark binary RAMnet. model against 23 other algorithms using real-world data from the StatLog Project. This large scale experimental study indicates that RAMnets are often capable of delivering results which are competitive with those obtained by more sophisticated, computationally expensive rnodels. The Frequency Weighted version is also benchmarked and shown to perform worse than the binary RAMnet for large values of the tuple size n. We demonstrate that the main issues in the Frequency Weighted RAMnets is adequate probability estimation and propose Good-Turing estimates in place of the more commonly used :Maximum Likelihood estimates. Having established the viability of the method numerically, we focus on providillg an analytical framework that allows us to quantify the generalisation cost of RAMnets for a given datasetL. For the classification network we provide a semi-quantitative argument which is based on the notion of Tuple distance. It gives a good indication of whether the network will fail for the given data. A rigorous Bayesian framework with Gaussian process prior assumptions is given for the regression n-tuple net. We show how to calculate the generalisation cost of this net and verify the results numerically for one dimensional noisy interpolation problems. We conclude that the n-tuple method of classification based on memorisation of random features can be a powerful alternative to slower cost driven models. The speed of the method is at the expense of its optimality. RAMnets will fail for certain datasets but the cases when they do so are relatively easy to determine with the analytical tools we provide.

Statistical physics of error-correcting codes

In this thesis we use statistical physics techniques to study the typical performance of four families of error-correcting codes based on very sparse linear transformations: Sourlas codes, Gallager codes, MacKay-Neal codes and Kanter-Saad codes. We map the decoding problem onto an Ising spin system with many-spins interactions. We then employ the replica method to calculate averages over the quenched disorder represented by the code constructions, the arbitrary messages and the random noise vectors. We find, as the noise level increases, a phase transition between successful decoding and failure phases. This phase transition coincides with upper bounds derived in the information theory literature in most of the cases. We connect the practical decoding algorithm known as probability propagation with the task of finding local minima of the related Bethe free-energy. We show that the practical decoding thresholds correspond to noise levels where suboptimal minima of the free-energy emerge. Simulations of practical decoding scenarios using probability propagation agree with theoretical predictions of the replica symmetric theory. The typical performance predicted by the thermodynamic phase transitions is shown to be attainable in computation times that grow exponentially with the system size. We use the insights obtained to design a method to calculate the performance and optimise parameters of the high performance codes proposed by Kanter and Saad.

Various techniques and procedures for refining ERS-1 orbits

The satellite ERS-1 was launched in July 1991 in a period of high solar activity. Sparse laser tracking and the failure of the experimental microwave system (PRARE) compounded the orbital errors which resulted from mismodelling of atmospheric density and hence surface forces. Three attempts are presented here to try and refine the coarse laser orbits of ERS-1, made prior to the availability of the full altimetric dataset. The results of the first attempt indicate that by geometrically modelling the satellite shape some improvement in orbital precision may be made for any satellite; especially one where no area tables already exist. The second and third refinement attempts are based on the introduction of data from some second satellite; in these examples SPOT-2 and TOPEX/Poseidon are employed. With SPOT-2 the method makes use of the orbital similarities to produce along-track corrections for the more fully tracked SPOT-2. Transferring these corrections to ERS-1 produces improvements in the precise orbits thus determined. With TOPEX/Poseidon the greater altitude results in a more precise orbit (gravity field and atmospheric errors are of less importance). Thus, by computing height differences at crossover points of the TOPEX/Poseidon and ERS-1 ground tracks the poorer orbit of ERS-1 may be improved by the addition of derived radial corrections. In the positive light of all three results several potential modification are suggested and some further avenues of investigation indicated.

Marketing strategy and new product development in the UK credit card industry

The development of new products in today's marketing environment is generally accepted as a requirement for the continual growth and prosperity of organisations. The literature is consequently rich with information on the development of various aspects of good products. In the case of service industries, it can be argued that new service product development is of as least equal importance as it is to organisations that produce tangible goods products. Unlike the new goods product literature, the literature on service marketing practices, and in particular, new service product development, is relatively sparse. The main purpose of this thesis is to examine a number of aspects of new service product development practice with respect to financial services and specifically, credit card financial services. The empirical investigation utilises both a case study and a survey approach, to examine aspects of new service product development industry practice relating specifically to gaps and deficiencies in the literature with respect to the financial service industry. The findings of the empirical work are subsequently examined in the context in which they provide guidance and support for a new normative new service product development model. The study examines the UK credit card financial service product sector as an industry case study and perspective. The findings of the field work reveal that the new service product development process is still evolving, and that in the case of credit card financial services can be seen as a well-structured and well-documented process. New product development can also be seen as an incremental, complex, interactive and continuous process which has been applied in a variety of ways. A number of inferences are subsequently presented.

Non-Linear oil firm force coefficients for a journal bearing operating under aligned and misaligned conditions

It is well established that hydrodynamic journal bearings are responsible for self-excited vibrations and have the effect of lowering the critical speeds of rotor systems. The forces within the oil film wedge, generated by the vibrating journal, may be represented by displacement and velocity coefficient~ thus allowing the dynamical behaviour of the rotor to be analysed both for stability purposes and for anticipating the response to unbalance. However, information describing these coefficients is sparse, misleading, and very often not applicable to industrial type bearings. Results of a combined analytical and experimental investigation into the hydrodynamic oil film coefficients operating in the laminar region are therefore presented, the analysis being applied to a 120 degree partial journal bearing having a 5.0 in diameter journal and a LID ratio of 1.0. The theoretical analysis shows that for this type of popular bearing, the eight linearized coefficients do not accurately describe the behaviour of the vibrating journal based on the theory of small perturbations, due to them being masked by the presence of nonlinearity. A method is developed using the second order terms of Taylor expansion whereby design charts are provided which predict the twentyeight force coefficients for both aligned, and for varying amounts of journal misalignment. The resulting non-linear equations of motion are solved using a modified Newton-Raphson method whereby the whirl trajectories are obtained, thus providing a physical appreciation of the bearing characteristics under dynamically loaded conditions.

The use of Meteosat satellite data for spatial rainfall estimations and hydrological simulations

Satellite information, in combination with conventional point source measurements, can be a valuable source of information. This thesis is devoted to the spatial estimation of areal rainfall over a region using both the measurements from a dense and sparse network of rain-gauges and images from the meteorological satellites. A primary concern is to study the effects of such satellite assisted rainfall estimates on the performance of rainfall-runoff models. Low-cost image processing systems and peripherals are used to process and manipulate the data. Both secondary as well as primary satellite images were used for analysis. The secondary data was obtained from the in-house satellite receiver and the primary data was obtained from an outside source. Ground truth data was obtained from the local Water Authority. A number of algorithms are presented that combine the satellite and conventional data sources to produce areal rainfall estimates and the results are compared with some of the more traditional methodologies. The results indicate that the satellite cloud information is valuable in the assessment of the spatial distribution of areal rainfall, for both half-hourly as well as daily estimates of rainfall. It is also demonstrated how the performance of the simple multiple regression rainfall-runoff model is improved when satellite cloud information is used as a separate input in addition to rainfall estimates from conventional means. The use of low-cost equipment, from image processing systems to satellite imagery, makes it possible for developing countries to introduce such systems in areas where the benefits are greatest.

Theoretical modelling of the sulcated spring

Replacement of the traditional coil spring with one of more fibre-reinforced plastic sulcated springs is a future possibility. Spring designers of metallic coil springs have design formulae readily available, and software packages specific to coil spring design exist. However, the sulcated spring is at the prototype stage of development, so literature on these springs is very sparse. The thesis contains information on the market for sulcated springs, and their advantages and disadvantages. Literature on other types of fibre reinforced plastic springs has also been reviewed. Design software has been developed for the sulcated spring along similar lines to coil spring design software. In order to develop the software, a theoretical model had to be developed which formed the mathematical basis for the software. The theoretical model is based on a choice of four methods for calculating the flexural rigidity; beam theory, plate theory, and lamination theory assuming isotropic and orthoropic material properties. Experimental results for strain and spring stiffness have been compared with the theoretical model, and were in good agreement. Included in the design software are the results of experimental work on fatigue, and design limiting factors to prevent or warn against impractical designs. Finite element analysis has been used to verify the theoretical model developed, and to find the better approximation to the experimental results. Applications and types of assemblies for the sulcated spring were discussed. Sulcated spring designs for the automotive applications of a suspension, clutch and engine valve spring were found using the design computer software. These sulcated spring designs were within or close to the space of the existing coil spring and yield the same performance. Finally the commercial feasibility of manufacturing the sulcated spring was assessed and compared with the coil spring, to evaluate the plausibility of the sulcated spring replacing the coil spring eventually.

Typical case behaviour of spin systems in random graph and composite ensembles

This thesis includes analysis of disordered spin ensembles corresponding to Exact Cover, a multi-access channel problem, and composite models combining sparse and dense interactions. The satisfiability problem in Exact Cover is addressed using a statistical analysis of a simple branch and bound algorithm. The algorithm can be formulated in the large system limit as a branching process, for which critical properties can be analysed. Far from the critical point a set of differential equations may be used to model the process, and these are solved by numerical integration and exact bounding methods. The multi-access channel problem is formulated as an equilibrium statistical physics problem for the case of bit transmission on a channel with power control and synchronisation. A sparse code division multiple access method is considered and the optimal detection properties are examined in typical case by use of the replica method, and compared to detection performance achieved by interactive decoding methods. These codes are found to have phenomena closely resembling the well-understood dense codes. The composite model is introduced as an abstraction of canonical sparse and dense disordered spin models. The model includes couplings due to both dense and sparse topologies simultaneously. The new type of codes are shown to outperform sparse and dense codes in some regimes both in optimal performance, and in performance achieved by iterative detection methods in finite systems.

Uncertainty and topographic visualisations

This thesis is a study of low-dimensional visualisation methods for data visualisation under certainty of the input data. It focuses on the two main feed-forward neural network algorithms which are NeuroScale and Generative Topographic Mapping (GTM) by trying to make both algorithms able to accommodate the uncertainty. The two models are shown not to work well under high levels of noise within the data and need to be modified. The modification of both models, NeuroScale and GTM, are verified by using synthetic data to show their ability to accommodate the noise. The thesis is interested in the controversy surrounding the non-uniqueness of predictive gene lists (PGL) of predicting prognosis outcome of breast cancer patients as available in DNA microarray experiments. Many of these studies have ignored the uncertainty issue resulting in random correlations of sparse model selection in high dimensional spaces. The visualisation techniques are used to confirm that the patients involved in such medical studies are intrinsically unclassifiable on the basis of provided PGL evidence. This additional category of ‘unclassifiable’ should be accommodated within medical decision support systems if serious errors and unnecessary adjuvant therapy are to be avoided.

Predictive gene lists for breast cancer prognosis: a topographic visualisation study

Background: The controversy surrounding the non-uniqueness of predictive gene lists (PGL) of small selected subsets of genes from very large potential candidates as available in DNA microarray experiments is now widely acknowledged 1. Many of these studies have focused on constructing discriminative semi-parametric models and as such are also subject to the issue of random correlations of sparse model selection in high dimensional spaces. In this work we outline a different approach based around an unsupervised patient-specific nonlinear topographic projection in predictive gene lists. Methods: We construct nonlinear topographic projection maps based on inter-patient gene-list relative dissimilarities. The Neuroscale, the Stochastic Neighbor Embedding(SNE) and the Locally Linear Embedding(LLE) techniques have been used to construct two-dimensional projective visualisation plots of 70 dimensional PGLs per patient, classifiers are also constructed to identify the prognosis indicator of each patient using the resulting projections from those visualisation techniques and investigate whether a-posteriori two prognosis groups are separable on the evidence of the gene lists. A literature-proposed predictive gene list for breast cancer is benchmarked against a separate gene list using the above methods. Generalisation ability is investigated by using the mapping capability of Neuroscale to visualise the follow-up study, but based on the projections derived from the original dataset. Results: The results indicate that small subsets of patient-specific PGLs have insufficient prognostic dissimilarity to permit a distinction between two prognosis patients. Uncertainty and diversity across multiple gene expressions prevents unambiguous or even confident patient grouping. Comparative projections across different PGLs provide similar results. Conclusion: The random correlation effect to an arbitrary outcome induced by small subset selection from very high dimensional interrelated gene expression profiles leads to an outcome with associated uncertainty. This continuum and uncertainty precludes any attempts at constructing discriminative classifiers. However a patient's gene expression profile could possibly be used in treatment planning, based on knowledge of other patients' responses. We conclude that many of the patients involved in such medical studies are intrinsically unclassifiable on the basis of provided PGL evidence. This additional category of 'unclassifiable' should be accommodated within medical decision support systems if serious errors and unnecessary adjuvant therapy are to be avoided.

From cardinal spline wavelet bases to highly coherent dictionaries

Wavelet families arise by scaling and translations of a prototype function, called the mother wavelet. The construction of wavelet bases for cardinal spline spaces is generally carried out within the multi-resolution analysis scheme. Thus, the usual way of increasing the dimension of the multi-resolution subspaces is by augmenting the scaling factor. We show here that, when working on a compact interval, the identical effect can be achieved without changing the wavelet scale but reducing the translation parameter. By such a procedure we generate a redundant frame, called a dictionary, spanning the same spaces as a wavelet basis but with wavelets of broader support. We characterize the correlation of the dictionary elements by measuring their 'coherence' and produce examples illustrating the relevance of highly coherent dictionaries to problems of sparse signal representation.

Palette-colouring: a belief propagation approach

We consider a variation of the prototype combinatorial optimization problem known as graph colouring. Our optimization goal is to colour the vertices of a graph with a fixed number of colours, in a way to maximize the number of different colours present in the set of nearest neighbours of each given vertex. This problem, which we pictorially call palette-colouring, has been recently addressed as a basic example of a problem arising in the context of distributed data storage. Even though it has not been proved to be NP-complete, random search algorithms find the problem hard to solve. Heuristics based on a naive belief propagation algorithm are observed to work quite well in certain conditions. In this paper, we build upon the mentioned result, working out the correct belief propagation algorithm, which needs to take into account the many-body nature of the constraints present in this problem. This method improves the naive belief propagation approach at the cost of increased computational effort. We also investigate the emergence of a satisfiable-to-unsatisfiable 'phase transition' as a function of the vertex mean degree, for different ensembles of sparse random graphs in the large size ('thermodynamic') limit.

Finding structures in information networks using the affinity network

This thesis proposes a novel graphical model for inference called the Affinity Network,which displays the closeness between pairs of variables and is an alternative to Bayesian Networks and Dependency Networks. The Affinity Network shares some similarities with Bayesian Networks and Dependency Networks but avoids their heuristic and stochastic graph construction algorithms by using a message passing scheme. A comparison with the above two instances of graphical models is given for sparse discrete and continuous medical data and data taken from the UCI machine learning repository. The experimental study reveals that the Affinity Network graphs tend to be more accurate on the basis of an exhaustive search with the small datasets. Moreover, the graph construction algorithm is faster than the other two methods with huge datasets. The Affinity Network is also applied to data produced by a synchronised system. A detailed analysis and numerical investigation into this dynamical system is provided and it is shown that the Affinity Network can be used to characterise its emergent behaviour even in the presence of noise.

Spatially extensive summation of contrast-energy is revealed by contrast detection of micro-pattern textures

Vision must analyze the retinal image over both small and large areas to represent fine-scale spatial details and extensive textures. The long-range neuronal convergence that this implies might lead us to expect that contrast sensitivity should improve markedly with the contrast area of the image. But this is at odds with the orthodox view that contrast sensitivity is determined merely by probability summation over local independent detectors. To address this puzzle, I aimed to assess the summation of luminance contrast without the confounding influence of area-dependent internal noise. I measured contrast detection thresholds for novel Battenberg stimuli that had identical overall dimensions (to clamp the aggregation of noise) but were constructed from either dense or sparse arrays of micro-patterns. The results unveiled a three-stage visual hierarchy of contrast summation involving (i) spatial filtering, (ii) long-range summation of coherent textures, and (iii) pooling across orthogonal textures. Linear summation over local energy detectors was spatially extensive (as much as 16 cycles) at Stage 2, but the resulting model is also consistent with earlier classical results of contrast summation (J. G. Robson & N. Graham, 1981), where co-aggregation of internal noise has obscured these long-range interactions.

Composite CDMA—a statistical mechanics analysis

Code division multiple access (CDMA) in which the spreading code assignment to users contains a random element has recently become a cornerstone of CDMA research. The random element in the construction is particularly attractive as it provides robustness and flexibility in utilizing multiaccess channels, whilst not making significant sacrifices in terms of transmission power. Random codes are generated from some ensemble; here we consider the possibility of combining two standard paradigms, sparsely and densely spread codes, in a single composite code ensemble. The composite code analysis includes a replica symmetric calculation of performance in the large system limit, and investigation of finite systems through a composite belief propagation algorithm. A variety of codes are examined with a focus on the high multi-access interference regime. We demonstrate scenarios both in the large size limit and for finite systems in which the composite code has typical performance exceeding those of sparse and dense codes at equivalent signal to noise ratio.