Studies on the generalisation of Gaussian processes and Bayesian neural networks

The assessment of the reliability of systems which learn from data is a key issue to investigate thoroughly before the actual application of information processing techniques to real-world problems. Over the recent years Gaussian processes and Bayesian neural networks have come to the fore and in this thesis their generalisation capabilities are analysed from theoretical and empirical perspectives. Upper and lower bounds on the learning curve of Gaussian processes are investigated in order to estimate the amount of data required to guarantee a certain level of generalisation performance. In this thesis we analyse the effects on the bounds and the learning curve induced by the smoothness of stochastic processes described by four different covariance functions. We also explain the early, linearly-decreasing behaviour of the curves and we investigate the asymptotic behaviour of the upper bounds. The effect of the noise and the characteristic lengthscale of the stochastic process on the tightness of the bounds are also discussed. The analysis is supported by several numerical simulations. The generalisation error of a Gaussian process is affected by the dimension of the input vector and may be decreased by input-variable reduction techniques. In conventional approaches to Gaussian process regression, the positive definite matrix estimating the distance between input points is often taken diagonal. In this thesis we show that a general distance matrix is able to estimate the effective dimensionality of the regression problem as well as to discover the linear transformation from the manifest variables to the hidden-feature space, with a significant reduction of the input dimension. Numerical simulations confirm the significant superiority of the general distance matrix with respect to the diagonal one.In the thesis we also present an empirical investigation of the generalisation errors of neural networks trained by two Bayesian algorithms, the Markov Chain Monte Carlo method and the evidence framework; the neural networks have been trained on the task of labelling segmented outdoor images.

Evolutionary neural networks : models and applications

The scaling problems which afflict attempts to optimise neural networks (NNs) with genetic algorithms (GAs) are disclosed. A novel GA-NN hybrid is introduced, based on the bumptree, a little-used connectionist model. As well as being computationally efficient, the bumptree is shown to be more amenable to genetic coding lthan other NN models. A hierarchical genetic coding scheme is developed for the bumptree and shown to have low redundancy, as well as being complete and closed with respect to the search space. When applied to optimising bumptree architectures for classification problems the GA discovers bumptrees which significantly out-perform those constructed using a standard algorithm. The fields of artificial life, control and robotics are identified as likely application areas for the evolutionary optimisation of NNs. An artificial life case-study is presented and discussed. Experiments are reported which show that the GA-bumptree is able to learn simulated pole balancing and car parking tasks using only limited environmental feedback. A simple modification of the fitness function allows the GA-bumptree to learn mappings which are multi-modal, such as robot arm inverse kinematics. The dynamics of the 'geographic speciation' selection model used by the GA-bumptree are investigated empirically and the convergence profile is introduced as an analytical tool. The relationships between the rate of genetic convergence and the phenomena of speciation, genetic drift and punctuated equilibrium arc discussed. The importance of genetic linkage to GA design is discussed and two new recombination operators arc introduced. The first, linkage mapped crossover (LMX) is shown to be a generalisation of existing crossover operators. LMX provides a new framework for incorporating prior knowledge into GAs.Its adaptive form, ALMX, is shown to be able to infer linkage relationships automatically during genetic search.

Study of the deposition, properties and applications of electroless deposits containing particles

The deposition efficiencies of a number of electroless nickel and cobalt plating solutions were studied and in the case of nickel compared with a commercial plating solution Nifoss 80. At the optimum plating conditions (92ºC and pH 4.5) Nifoss 80 produced nickel layers most efficiently, the alkaline cobalt solution operated most efficiently at 90ºC and pH 9. The methods of producing compostte layers containing 2-3 µm carbide particles and chromium powder is described. Nickel and cobalt layers containing approximately 27% carbide particles, or 40% (Ni) and 30% (Co) chromium particles by volume were obtained. This value is independent of the particle concentration in the plating solution within the range (20~200 g/l). Hardness of the nickel. as deposited was 515 Hv, this was increased to a maximum of 1155 Hv by heat treatment at 200ºC for 5 hours in vacuum. Incorporation. of .chromium carbide particles resulted in a maximum hardness of 1225 Hv after heating at 500ºC for 5 hours in vacuum and chromium particles resulted in a maximum hardness of 16S0 Hv after heat treatment at 400ºC for 2 hours in vacuum. Similarly the hardness of cobalt as deposited was 600 Hv, this was increased to a maximum of 1300 Hv after heat treatment at 400ºC for 1 hour. Incorporation of chromium carbide particles resulted jn a maximum hardness of 1405 Hv after heating at 400ºC for 5 hours in vacuum and chromium particles resulted in a maximum hardness of 1440 Hv after. heat treating for 2 hours at 400ºC in vacuum. The structure of the deposits was studied by optical and scanning electron microscopy. The wear rate and coefficient of friction was determined by a pin and disc method. Wear rate and coefficient of friction decreased with increase in hardness. The wear resistance of the materials was also determined using a simulated forging test. Dies made of standard die steel were coated and the wear rates of the layers as deposited and after heat treatment were compared with those of uncoated tools. The wear resistance generally increased with hardness, it was 50-75% more than the uncoated die steel. Acetic acid salt spray test and outdoor exposure for six months was used to study the corrosion behaviour of the deposits and potentiodynamic curves plotted to find their corrosion potential. Nickel deposit exhibited less staining than carbide composite deposits and nickel-chromium deposits had the most noble corrosion potential.

The impact of visual-verbal relationships on native-nonnative english speakers' reading processes and comprehension

The aim of this study was to comparatively investigate the impact of visual-verbal relationships that exist in expository texts on the reading process and comprehension of readers from different language background: native speakers of English (LI) and speakers of English as a foreign language (EFL). The study focussed, in this respect, on the visual elements (VEs) mainly graphs and tables that accompanied the selected texts. Two major experiments were undertaken. The first, was for the reading process using the post-reading questionnaire technique. Participants were 163 adult readers representing three groups: 77 (LI), 56 (EFL postgraduates); and 30 (EFL undergraduates). The second experiment was for the reading comprehension using cloze procedure. Participants were 123 representing the same above gorups: 50, 33 and 40 respectively. It was hypothesised that the LI readers would make use of VEs in the reading process in ways different from both EFL groups and that use would enhance each group's comprehension in different aspects and to different levels. In the analysis of the data of both experiments two statistical measurements were used. The chi-square was used to measure the differences between frequencies and the t-test was used to measure the differences between means. The results indicated a significant relationship between readers' language background and the impact of visual-verbal relationships on their reading processes and comprehension of such type of texts. The results also revealed considerable similarities between the two EFL groups in the reading process of texts accompanied by VEs. In the reading comprehension, however, the EFL undergraduates seemed to benefit from the visual-verbal relationships in their comprehension more than the postgraduates, suggesting a weak relationship of this impact for older EFL readers. Furthermore, the results showed considerable similarities between the reading process of texts accompanied by VEs and of whole prose texts. Finally an evaluation of this study was undertaken as well as practical implications for EFL readers and suggestions for future research.

UV-inscription, polarization-dependant loss characteristics and applications of 45° tilted fiber gratings

We have UV-inscribed and theoretically and experimentally analyzed fiber gratings with the structure tilted at 45° and implemented this type of devices as an in-fiber polarizer. A systematic investigation has been carried out on the characterization of 45° tilted fiber gratings (45° TFGs) in terms of the polarization-dependant loss (PDL) and thermal response. The detailed theoretical modeling has revealed a linear correlation between the grating length and the PDL, which has been proved by the experimental results. For the first time, we have examined the UV beam diffraction from a tilted phase mask and designed the UV-inscription system to suit the 45° TFG fabrication. Experimentally, a 24 mm long 45° TFG UV-inscribed in standard telecom single-mode fiber exhibited around 25 dB PDL at 1530 nm and an over ~300 nm bandwidth of PDL spectrum. By the concatenation method, a 44 mm long grating showed a PDL as high as 40 dB that is close to the high polarization extinction ratio of commercial products. Moreover, we have revealed that the PDL of 45° TFGs has low thermal influence, which is desirable for real application devices. Finally, we experimentally demonstrated an all-fiber twist sensor system based on a 45° and an 81° TFG.

Survey and applications of standardized energy metrics to mobile networks

The energy consumption and the energy efficiency have become very important issue in optimizing the current as well as in designing the future telecommunications networks. The energy and power metrics are being introduced in order to enable assessment and comparison of the energy consumption and power efficiency of the telecommunications networks and other transmission equipment. The standardization of the energy and power metrics is a significant ongoing activity aiming to define the baseline energy and power metrics for the telecommunications systems. This article provides an up-to-date overview of the energy and power metrics being proposed by the various standardization bodies and subsequently adopted worldwide by the equipment manufacturers and the network operators. © Institut Télécom and Springer-Verlag 2012.and Springer-Verlag 2012.

Measurement of implicit leadership theories and their effect on leadership processes and outcomes

The leadership categorisation theory suggests that followers rely on a hierarchical cognitive structure in perceiving leaders and the leadership process, which consists of three levels; superordinate, basic and subordinate. The predominant view is that followers rely on Implicit Leadership Theories (ILTs) at the basic level in making judgments about managers. The thesis examines whether this presumption is true by proposing and testing two competing conceptualisations; namely the congruence between the basic level ILTs (general leader) and actual manager perceptions, and subordinate level ILTs (job-specific leader) and actual manager. The conceptualisation at the job-specific level builds on context-related assertions of the ILT explanatory models: leadership categorisation, information processing and connectionist network theories. Further, the thesis addresses the effects of ILT congruence at the group level. The hypothesised model suggests that Leader-Member Exchange (LMX) will act as a mediator between ILT congruence and outcomes. Three studies examined the proposed model. The first was cross-sectional with 175 students reporting on work experience during a 1-year industrial placement. The second was longitudinal and had a sample of 343 students engaging in a business simulation in groups with formal leadership. The final study was a cross-sectional survey in several organisations with a sample of 178. A novel approach was taken to congruence analysis; the hypothesised models were tested using Latent Congruence Modelling (LCM), which accounts for measurement error and overcomes the majority of limitations of traditional approaches. The first two studies confirm the traditional theorised view that employees rely on basic-level ILTs in making judgments about their managers with important implications, and show that LMX mediates the relationship between ILT congruence and work-related outcomes (performance, job satisfaction, well-being, task satisfaction, intragroup conflict, group satisfaction, team realness, team-member exchange, group performance). The third study confirms this with conflict, well-being, self-rated performance and commitment as outcomes.

Minority influence and innovation:antecedents, processes and consequences

Social groups form an important part of our daily lives. Within these groups pressures exist which encourage the individual to comply with the group’s viewpoint. This influence, which creates social conformity, is known as ‘majority influence’ and is the dominant process of social control. However, there also exists a ‘minority influence’, which emerges from a small subsection of the group and is a dynamic force for social change. Minority Influence and Innovation seeks to identify the conditions under which minority influence can prevail, to change established norms, stimulate original thinking and help us to see the world in new ways. With chapters written by a range of expert contributors, areas of discussion include: •processes and theoretical issues •the factors which affect majority and minority influence •interactions between majority and minority group members This book offers a thorough evaluation of the most important current developments within this field and presents consideration of the issues that will be at the forefront of future research. As such it will be of interest to theorists and practitioners working in social psychology. This book offers a thorough evaluation of the most important current developments within this field and presents consideration of the issues that will be at the forefront of future research. As such it will be of interest to theorists and practitioners working in social psychology.

Solvability and asymptotics of the heat equation with mixed variable lateral conditions and applications in the opening of the exocytotic fusion pore in cells

We investigate a mixed problem with variable lateral conditions for the heat equation that arises in modelling exocytosis, i.e. the opening of a cell boundary in specific biological species for the release of certain molecules to the exterior of the cell. The Dirichlet condition is imposed on a surface patch of the boundary and this patch is occupying a larger part of the boundary as time increases modelling where the cell is opening (the fusion pore), and on the remaining part, a zero Neumann condition is imposed (no molecules can cross this boundary). Uniform concentration is assumed at the initial time. We introduce a weak formulation of this problem and show that there is a unique weak solution. Moreover, we give an asymptotic expansion for the behaviour of the solution near the opening point and for small values in time. We also give an integral equation for the numerical construction of the leading term in this expansion.