The theory of on-line learning: a statistical physics approach

In this paper we review recent theoretical approaches for analysing the dynamics of on-line learning in multilayer neural networks using methods adopted from statistical physics. The analysis is based on monitoring a set of macroscopic variables from which the generalisation error can be calculated. A closed set of dynamical equations for the macroscopic variables is derived analytically and solved numerically. The theoretical framework is then employed for defining optimal learning parameters and for analysing the incorporation of second order information into the learning process using natural gradient descent and matrix-momentum based methods. We will also briefly explain an extension of the original framework for analysing the case where training examples are sampled with repetition.

Smart microsystems for cell manipulations

This thesis documents the design, manufacture and testing of a passive and non-invasive micro-scale planar particle-from-fluid filter for segregating cell types from a homogeneous suspension. The microfluidics system can be used to separate spermatogenic cells from testis biopsy samples, providing a mechanism for filtrate retrieval for assisted reproduction therapy. The system can also be used for point-of-service diagnostics applications for hospitals, lab-on-a-chip pre-processing and field applications such as clinical testing in the third world. Various design concepts are developed and manufactured, and are assessed based on etched structure morphology, robustness to variations in the manufacturing process, and design impacts on fluid flow and particle separation characteristics. Segregation was measured using image processing algorithms that demonstrate efficiency is more than 55% for 1 µl volumes at populations exceeding 1 x 107. the technique supports a significant reduction in time over conventional processing, in the separation and identification of particle groups, offering a potential reduction in the associated cost of the targeted procedure. The thesis has developed a model of quasi-steady wetting flow within the micro channel and identifies the forces across the system during post-wetting equalisation. The model and its underlying assumptions are validated empirically in microfabricated test structures through a novel Micro-Particle Image Velocimetry technique. The prototype devices do not require ancillary equipment nor additional filtration media, and therefore offer fewer opportunities for sample contamination over conventional processing methods. The devices are disposable with minimal reagent volumes and process waste. Optimal processing parameters and production methods are identified with any improvements that could be made to enhance their performance in a number of identified potential applications.

Improving the repeatability of heterochromatic flicker photometry for measurement of macular pigment optical density

Background: Heterochromatic flicker photometry (HFP) is a psychophysical technique used to measure macular pigment optical density (MPOD). We used the MPS 9000 (MPS) HFP device. Our aim was to determine if the repeatability of the MPS could be improved to make it more suitable for monitoring MPOD over time. Methods: Intra-session repeatability was assessed in 25 participants (aged 20-50 years). The resulting data was explored in detail, e.g., by examining the effect of removal and adjustment of data with less than optimal quality parameters. A protocol was developed for improved overall reliability, which was then tested in terms of inter-session repeatability in a separate group of 27 participants (aged 19-52 years). Results: Removal and adjustment of data reduced the intra-session coefficient of repeatability (CR) by 0.04, on average, and the mean individual standard deviation by 0.004. Raw data observation offered further insight into ways of improving repeatability. The proposed protocol resulted in an inter-session CR of 0.08. Conclusions: Removal and adjustment of less than optimal data improved repeatability, and is therefore recommended. To further improve repeatability, in brief we propose that each patient perform each part of the test twice, and a third time where necessary (described in detail by the protocol). Doing so will make the MPS more useful in research and clinical settings. © 2012 Springer-Verlag.

Numerical investigation of the effect of the temporal pulse shape on modification of fused silica by femtosecond pulses

We report the results of numerical studies of the impact of asymmetric femtosecond pulses focused in the bulk of the material on the femtosecond modification of fused silica. It is shown that such pulses lead to localisation of absorption in the process of femtosecond modification and to a decrease in the threshold energy of modification. It is found that the optimal asymmetry parameters for reaching the maximum plasma density in the focusing region depend on the pulse energy: at an initial energy of about 100 nJ, it is preferable to use pulses with positive TOD; however, when the energy is increased, it is preferable to use pulses with negative TOD. This is explained by differences in the dynamics of the processes of absorption of energy of a pulse propagating in the material.

Optimization of discrete Raman amplifiers for different kinds of fibers

We present an analysis of the performance of backward-pumped discrete Raman amplifier modules designed for simultaneous amplification and dispersion and/or dispersion slope compensation, both in single-channel and in multichannel systems. Optimal module parameters are determined within a realistic range of pump and signal powers.

Globally optimal parameters for on-line learning in multilayer neural networks

We present a framework for calculating globally optimal parameters, within a given time frame, for on-line learning in multilayer neural networks. We demonstrate the capability of this method by computing optimal learning rates in typical learning scenarios. A similar treatment allows one to determine the relevance of related training algorithms based on modifications to the basic gradient descent rule as well as to compare different training methods.

The factors affecting the optimal performance of eddy-current machines

This thesis presents an examination of the factors which influence the performance of eddy-current machines and the way in which they affect optimality of those machines. After a brief introduction to the types of eddy-current machine considered, the applications to which these machines are put are examined. A list of parameters by which to assess their performance is obtained by considering the machine as part of a system. in this way an idea of what constitutes an optimal machine is obtained. The third chapter then identifies the factors which affects the performance and makes a quantitative evaluation of the effect. Here the various alternative configurations and components are compared with regard to their influence on the mechanical, electromagnetic, and thermal performance criteria of the machine. Chapter four contains a brief review of the methods of controlling eddy-current machines by electronic methods using thyristors or transistors as the final control element. Where necessary, the results of previous workers in the field of electrical machines have been extended or adapted to increase the usefulness of this thesis.

The role of stochasticity in an information-optimal neural population code

In this paper we consider the optimisation of Shannon mutual information (MI) in the context of two model neural systems The first is a stochastic pooling network (population) of McCulloch-Pitts (MP) type neurons (logical threshold units) subject to stochastic forcing; the second is (in a rate coding paradigm) a population of neurons that each displays Poisson statistics (the so called 'Poisson neuron'). The mutual information is optimised as a function of a parameter that characterises the 'noise level'-in the MP array this parameter is the standard deviation of the noise, in the population of Poisson neurons it is the window length used to determine the spike count. In both systems we find that the emergent neural architecture and; hence, code that maximises the MI is strongly influenced by the noise level. Low noise levels leads to a heterogeneous distribution of neural parameters (diversity), whereas, medium to high noise levels result in the clustering of neural parameters into distinct groups that can be interpreted as subpopulations In both cases the number of subpopulations increases with a decrease in noise level. Our results suggest that subpopulations are a generic feature of an information optimal neural population.

The Kinematics of interference mechanisms in certain machining operations

The mechanism of "Helical Interference" in milled slots is examined and a coherent theory for the geometry of such surfaces is presented. An examination of the relevant literature shows a fragmented approach to the problem owing to its normally destructive nature, so a complete analysis is developed for slots of constant lead, thus giving a united and exact theory for many different setting parameters and a range of cutter shapes. For the first time, a theory is developed to explain the "Interference Surface" generated in variable lead slots for cylindrical work and attention is drawn to other practical surfaces, such as cones, where variable leads are encountered. Although generally outside the scope of this work, an introductory analysis of these cases is considered in order to develop the cylindrical theory. Special emphasis is laid upon practical areas where the interference mechanism can be used constructively and its application as the rake face of a cutting tool is discussed. A theory of rake angle for such cutting tools is given for commonly used planes, and relative variations in calculated rake angle between planes is examined. Practical tests are conducted to validate both constant lead and variable lead theories and some design improvements to the conventional dividing head are suggested in order to manufacture variable lead workpieces, by use of a "superposed" rotation. A prototype machine is manufactured and its kinematic principle given for both linear and non-linearly varying superposed rotations. Practical workpieces of the former type are manufactured and compared with analytical predictions,while theoretical curves are generated for non-linear workpieces and then compared with those of linear geometry. Finally suggestions are made for the application of these principles to the manufacture of spiral bevel gears, using the "Interference Surface" along a cone as the tooth form.

Correlations between optimal temporal width and spectral characteristics of an optical signal in a wavelength-paired CS-RZ transmission with high spectral efficiency

We examine the correlations between the parameters of ultra-narrow off-centred filtering and pulse width on the performance of a wavelength paired Nx40Gbit/s DWDM transmission, consisting of carrier suppressed return-to-zero signal with 0.64 bit/s/Hz (without polarization-division multiplexing) spectral efficiency.

A hierarchy of phase transitions in optimal neuronal coding:from binary to M-ary discrete optimal codes

We have investigated how optimal coding for neural systems changes with the time available for decoding. Optimization was in terms of maximizing information transmission. We have estimated the parameters for Poisson neurons that optimize Shannon transinformation with the assumption of rate coding. We observed a hierarchy of phase transitions from binary coding, for small decoding times, toward discrete (M-ary) coding with two, three and more quantization levels for larger decoding times. We postulate that the presence of subpopulations with specific neural characteristics could be a signiture of an optimal population coding scheme and we use the mammalian auditory system as an example.

Correlations between optimal temporal width and spectral characteristics of an optical signal in a wavelength-paired CS-RZ transmission with high spectral efficiency

We examine the correlations between the parameters of ultra-narrow off-centred filtering and pulse width on the performance of a wavelength paired Nx40Gbit/s DWDM transmission, consisting of carrier suppressed return-to-zero signal with 0.64 bit/s/Hz (without polarization-division multiplexing) spectral efficiency. © 2004 Optical Society of America.

Manufacturing process challenges when machining superalloys

The implementation of advanced manufacturing systems with high process capability is an essential requirement for the high value manufacturing industries. To ensure high process capability, industry needs to deal with the requirement for tight tolerances and the unavoidable variations in materials, and manufacturing and inspection processes. In the case of machining superalloys, such variations result in the need to change the machine parameters for producing different batches of materials from different suppliers. This is required in order to get the process under control and reduce waste and defects, leading to better competitiveness. This papers discuss the variability in materials and the corresponding process requirements when machining superalloys, and highlights the impact of metrology in achieving manufacturing process improvement.

Integrated optimisation of surface roughness and tool performance when face milling 416 SS

Tool life is an important factor to be considered during the optimisation of a machining process since cutting parameters can be adjusted to optimise tool changing, reducing cost and time of production. Also the performance of a tool is directly linked to the generated surface roughness and this is important in cases where there are strict surface quality requirements. The prediction of tool life and the resulting surface roughness in milling operations has attracted considerable research efforts. The research reported herein is focused on defining the influence of milling cutting parameters such as cutting speed, feed rate and axial depth of cut, on three major tool performance parameters namely, tool life, material removal and surface roughness. The research is seeking to define methods that will allow the selection of optimal parameters for best tool performance when face milling 416 stainless steel bars. For this study the Taguchi method was applied in a special design of an orthogonal array that allows studying the entire parameter space with only a number of experiments representing savings in cost and time of experiments. The findings were that the cutting speed has the most influence on tool life and surface roughness and very limited influence on material removal. By last tool life can be judged either from tool life or volume of material removal.