Gaussian processes:iterative sparse approximations

In recent years there has been an increased interest in applying non-parametric methods to real-world problems. Significant research has been devoted to Gaussian processes (GPs) due to their increased flexibility when compared with parametric models. These methods use Bayesian learning, which generally leads to analytically intractable posteriors. This thesis proposes a two-step solution to construct a probabilistic approximation to the posterior. In the first step we adapt the Bayesian online learning to GPs: the final approximation to the posterior is the result of propagating the first and second moments of intermediate posteriors obtained by combining a new example with the previous approximation. The propagation of em functional forms is solved by showing the existence of a parametrisation to posterior moments that uses combinations of the kernel function at the training points, transforming the Bayesian online learning of functions into a parametric formulation. The drawback is the prohibitive quadratic scaling of the number of parameters with the size of the data, making the method inapplicable to large datasets. The second step solves the problem of the exploding parameter size and makes GPs applicable to arbitrarily large datasets. The approximation is based on a measure of distance between two GPs, the KL-divergence between GPs. This second approximation is with a constrained GP in which only a small subset of the whole training dataset is used to represent the GP. This subset is called the em Basis Vector, or BV set and the resulting GP is a sparse approximation to the true posterior. As this sparsity is based on the KL-minimisation, it is probabilistic and independent of the way the posterior approximation from the first step is obtained. We combine the sparse approximation with an extension to the Bayesian online algorithm that allows multiple iterations for each input and thus approximating a batch solution. The resulting sparse learning algorithm is a generic one: for different problems we only change the likelihood. The algorithm is applied to a variety of problems and we examine its performance both on more classical regression and classification tasks and to the data-assimilation and a simple density estimation problems.

WDM of solitons in dispersion managed transmission systems

This thesis investigates the physical behaviour of solitons in wavelength division multiplexed (WDM) systems with dispersion management in a wide range of dispersion regimes. Background material is presented to show how solitons propagate in optical fibres, and key problems associated with real systems are outlined. Problems due to collision induced frequency shifts are calculated using numerical simulation, and these results compared with analytical techniques where possible. Different two-step dispersion regimes, as well as the special cases of uniform and exponentially profiled systems, are identified and investigated. In shallow profile, the constituent second-order dispersions in the system are always close to the average soliton value. It is shown that collision-induced frequency shifts in WDM soliton transmission systems are reduced with increasing dispersion management. New resonances in the collision dynamics are illustrated, due to the relative motion induced by the dispersion map. Consideration of third-order dispersion is shown to modify the effects of collision-induced timing jitter and third-order compensation investigated. In all cases pseudo-phase-matched four-wave mixing was found to be insignificant compared to collision induced frequency shift in causing deterioration of data. It is also demonstrated that all these effects are additive with that of Gordon-Haus jitter.

2R and 3R optical regeneration in 40 Gbit/s WDM terrestrial networks

We analyse a 2R regenerator using nonlinear-optical-loop-mirror and a 3R regenerator employing nonlinearly-enhanced amplitude modulator in 40Gbit/s WDM networks based on standard fibre (SMF). Characterization of one- (600km of SMF) and two-step regeneration is presented.

Kinetic study on thermal decomposition of woods in oxidative environment

The purpose of this work is to gain knowledge on kinetics of biomass decomposition under oxidative atmospheres, mainly examining effect of heating rate on different biomass species. Two sets of experiments are carried out: the first set of experiments is thermal decomposition of four different wood particles, namely aspens, birch, oak and pine under an oxidative atmosphere and analysis with TGA; and the second set is to use large size samples of wood under different heat fluxes in a purpose-built furnace, where the temperature distribution, mass loss and ignition characteristics are recorded and analyzed by a data post-processing system. The experimental data is then used to develop a two-step reactions kinetic scheme with low and high temperature regions while the activation energy for the reactions of the species under different heating rates is calculated. It is found that the activation energy of the second stage reaction for the species with similar constituent fractions tends to converge to a similar value under the high heating rate.

Environmental degradation of polyethylene-based plastics

The criteria involved in the degradation of polyethylene-based degradable polymer samples have been investigated, with a view to obtaining a clearer mechanism of photo-biodegradation. The compatibility of degradable polymer samples during materials recycling was also studied. Commercial and laboratory prepared degradable polymer samples were oxidised in different environments and the oxidation products formed were studied using various analytical chromatographic and spectroscopic techniques such as HPLC, FT-IR and NMR. It was found that commercial degradable polymer samples which are based on the ECO systems, degrade predominantly via the Norrish II process, whereas the other degradable systems studied (starch-filled polyethylene systems, transition metal systems, including metal carboxylate based polyethylene systems and the photoantioxidant-activator systems) photodegrade essentially via the Norrish I process. In all cases, the major photoxidation products extracted from the degradable polymer samples were found to be carboxylic acids, although, in the polymer itself a mixture of carbonyl containing products such as esters, lactones, ketones and aldehydes was observed. The study also found that the formation of these hydrophilic carbonyl products causes surface swelling of the polymer, thus making bioerosion possible. It was thus concluded that environmental degradation of LDPE is a two step process, the initiation stage being oxidation of the polymer which gives rise to bioassimilable products, which are consequently bioeroded in the second stage, (the biodegradation step). Recycling of the degradable polymer samples as 10% homogeneous and heterogeneous blends was carried out using a single screw extruder (180°C and 210°C) and an internal mixer (190°C). The study showed that commercial degradable polymer samples may be recycled with a minimal loss in their properties.

Enterotoxigenic Escherichia coli infection of pigs: expression, extraction, purification and the adhesive properties of the K88 fimbrial adhesin

The ability of Escherichia coli to express the K88 fimbrial adhesin was satisfactorily indicated by the combined techniques of ELISA, haemagglutination and latex agglutination. Detection of expression by electron microscopy and the ability to metabolize raffinose were unsuitable. Quantitative expression of the K88 adhesin was determined by ELISA. Expression was found to vary according to the E.coli strain examined, media type and form. In general it was found that the total amount was greater, while the amount/cfu was less on agar than in broth cultures. Expression of the K88 adhesin during unshaken batch culture was related to the growth rate and was maximal during late logarithmic to early stationary phase. A combination of heat extraction, ammonium sulphate and isoelectric precipitation was found suitable for both large and small scale preparation of purified K88ab adhesin. Extraction of the K88 adhesin was sensitive to pH and it was postulated that this may affect the site of colonisation of by ETEC in vivo. Results of haemagglutination experiments were consistent with the hypothesis that the K88 receptor present on erythrocytes is composed of two elements, one responsible for the binding of K88ab and K88ac and a second responsible for the binding of the K88ad adhesin. Comparison of the haemagglutinating properties of cell-free and cell-bound K88 adhesin revealed some differences probably indicating a minor conformational change in the K88 adhesin on its isolation. The K88ab adhesin was found to bind to erythrocytes over a wide pH range (PH 4-9) and was inhibited by αK88ab and αK88b antisera. Inhibition of haemagglutination was noted with crude heparin, mannan and porcine gastric mucin, chondrosine and several hexosamines, glucosamine in particular. The most potent inhibitor of haemagglutination was n-dodecyl-β-D-glucopyranoside, one of a series of glucosides found to have inhibitory properties. Correlation between hydrophobicity of glucosides tested and degree of inhibition observed suggested hydrophobic forces were important in the interaction of the K88 adhesin with its receptor. The results of Scatchard and Hill plots indicated that binding of the K88ab adhesin to porcine enterocytes in the majority of cases is a two-step, three component system. The first K88 receptor (or site) had a K2. of 1.59x1014M-1 and a minimum of 4.3x104 sites/enterocyte. The second receptor (or site) had a K2 of 4.2x1012M-1 with a calculated 1.75x105 sites/enterocyte. Attempts to inhibit binding of cell-free K88 adhesin to porcine enterocytes by lectins were unsuccessful. However, several carbohydrates including trehalose, lactulose, galactose 1→4 mannopyranoside, chondrosine, galactosamine, stachyose and mannan were inhibitory. The most potent inhibitor was found to be porcine gastric mucin. Inhibition observed with n-octyl-α-D-glucopyranose was difficult to interpret in isolation because of interference with the assay, however, it agreed with the results of haemagglutination inhibition experiments.

Contract specificity and its performance outcomes

Governance theories, such as transaction cost economics, argue that systematic deviations from an attribute–governance alignment should influence performance. This article investigates the performance implications of contract specificity for the procurement of information technology products. The authors argue that parties choose a level of contract specificity that economizes on both the ex ante contracting costs and the ex post transaction costs and that deviations between the observed and the predicted levels of contract specificity are an important determinant of these transaction costs. The authors test the hypotheses using a comprehensive archival data set of information technology transactions and employ a two-step estimation procedure. First, they estimate the “predicted” level of contract specificity, which accounts for key transactional attributes. Second, they study the consequences of deviating from this predicted level of contractual specificity. The results provide the first explicit demonstration of the trade-off between ex ante contracting costs and ex post transaction problems and suggest that parties need to economize jointly on these costs when choosing the governance form.

In-fiber microchannel device filled with a carbon nanotube dispersion for passive mode-lock lasing

Fueled by their high third-order nonlinearity and nonlinear saturable absorption, carbon nanotubes (CNT) are expected to become an integral part of next-generation photonic devices such as all-optical switches and passive mode-locked lasers. However, in order to fulfill this expectation it is necessary to identify a suitable platform that allows the efficient use of the optical properties of CNT. In this paper, we propose and implement a novel device consisting of an optofluidic device filled with a dispersion of CNT. By fabricating a microchannel through the core of a conventional fiber and filling it with a homogeneous solution of CNTs on Dimethylformamide (DMF), a compact, all-fiber saturable absorber is realized. The fabrication of the micro-fluidic channel is a two-step process that involves femtosecond laser micro-fabrication and chemical etching of the laser-modified regions. All-fiber high-energy, passive mode-locked lasing is demonstrated with an output power of 13.5 dBm. The key characteristics of the device are compactness and robustness against optical, mechanical and thermal damage.

Performance comparison of 2R and 3R optical regeneration schemes at 40 Gb/s for application to all-optical networks

This work numerically analyzes the performances of a 2R (reamplification and reshaping) regenerator based on a nonlinear optical loop mirror and a 3R (reamplification, reshaping, and retiming) regenerator using a nonlinearly enhanced amplitude modulator in 40-Gb/s standard single-mode fiber (SMF)-based optical networks with large amplifier spacing. The characteristics of one(600 km of SMF) and two-step regeneration are examined and the feasibility of wavelength-division multiplexing (WDM) operation is demonstrated.

"Magic" dispersion maps with distributed Raman amplification

We analyze pulse propagation in an optical fiber with a periodic dispersion map and distributed amplification. Using an asymptotic theory and a momentum method, we identify a family of dispersion management schemes that are advantageous for massive multichannel soliton transmission. For the case of two-step dispersion maps with distributed Raman amplification to compensate for the fiber loss, we find special schemes that have optimal (chirp-free) launch point locations that are independent of the fiber dispersion. Despite the variation of dispersion with wavelength due to the fiber dispersion slope, the transmission in several different channels can be optimized simultaneously using the same optimal launch point. The theoretical predictions are verified by direct numerical simulations. The obtained results are applied to a practical multichannel transmission system.

Optimal dispersion maps for multichannel soliton transmission with distributed Raman amplification

Using an asymptotic theory and a momentum method, we identify a family of dispersion management schemes with distributed Raman amplification, which are advantageous for massive multichannel soliton transmission. For the case of two-step dispersion maps, special schemes are found that have optimal (chirp-free) launch point locations that are independent of the fibre dispersion. Despite the variation of dispersion with wavelength due to the fibre dispersion slope, the transmission in several different channels can be optimized simultaneously using the same optimal launch point. The theoretical results are verified by direct numerical simulations.

Economic efficiency of smallholder maize producers in Western Kenya:a DEA meta-frontier analysis

Maize is the main staple food for most Kenyan households, and it predominates where smallholder, as well as large-scale, farming takes place. In the sugarcane growing areas of Western Kenya, there is pressure on farmers on whether to grow food crops, or grow sugarcane, which is the main cash crop. Further, with small and diminishing land sizes, the question of productivity and efficiency, both for cash and food crops is of great importance. This paper, therefore, uses a two-step estimation technique (DEA meta-frontier and Tobit Regression) to highlight the inefficiencies in maize cultivation, and their causes in Western Kenya.

2R and 3R optical regeneration in 40 Gbit/s WDM terrestrial networks

We analyse a 2R regenerator using nonlinear-optical-loop-mirror and a 3R regenerator employing nonlinearly-enhanced amplitude modulator in 40Gbit/s WDM networks based on standard fibre (SMF). Characterization of one- (600km of SMF) and two-step regeneration is presented.

Performance comparison of 2R and 3R optical regeneration schemes at 40 Gb/s for application to all-optical networks

This paper numerically analyzes the performances of a 2R (reamplification and reshaping) regenerator based on a nonlinear optical loop mirror and a 3R (reamplification, reshaping, and retiming) regenerator using a nonlinearly enhanced amplitude modulator in 40-Gb/s standard single-mode fiber (SMF)-based optical networks with large amplifier spacing. The characteristics of one- (600 km of SMF) and two-step regeneration are examined and the feasibility of wavelength-division multiplexing operation is demonstrated. © 2005 IEEE.

Structure and physical properties of [µ-Tris(1,4-bis(tetrazol-1-yl)butane-N4,N4‘')iron(II)] Bis(hexafluorophosphate), a new Fe(II) spin-crossover compound with a three-dimensional threefold interlocked crystal lattice

[μ-Tris(1,4-bis(tetrazol-1-yl)butane-N4,N4‘)iron(II)] bis(hexafluorophosphate), [Fe(btzb)3](PF6)2, crystallizes in a three-dimensional 3-fold interlocked structure featuring a sharp two-step spin-crossover behavior. The spin conversion takes place between 164 and 182 K showing a discontinuity at about T1/2 = 174 K and a hysteresis of about 4 K between T1/2 and the low-spin state. The spin transition has been independently followed by magnetic susceptibility measurements, 57Fe-Mössbauer spectroscopy, and variable temperature far and midrange FTIR spectroscopy. The title compound crystallizes in the trigonal space group P30¯(No. 147) with a unit cell content of one formula unit plus a small amount of disordered solvent. The lattice parameters were determined by X-ray diffraction at several temperatures between 100 and 300 K. Complete crystal structures were resolved for 9 of these temperatures between 100 (only low spin, LS) and 300 K (only high spin, HS), Z = 1 [Fe(btzb)3](PF  6)2:  300 K (HS), a = 11.258(6) Å, c = 8.948(6) Å, V = 982.2(10) Å3; 100 K (LS), a = 10.989(3) Å, c = 8.702(2) Å, V = 910.1(4) Å3. The molecular structure consists of octahedral coordinated iron(II) centers bridged by six N4,N4‘ coordinating bis(tetrazole) ligands to form three 3-dimensional networks. Each of these three networks is symmetry related and interpenetrates each other within a unit cell to form the interlocked structure. The Fe−N bond lengths change between 1.993(1) Å at 100 K in the LS state and 2.193(2) Å at 300 K in the HS state. The nearest Fe separation is along the c-axis and identical with the lattice parameter c.