For many types of learners one can compute the statistically 'optimal' way to select data. We review how these techniques have been used with feedforward neural networks. We then show how the same principles may be used to select data for two alternative, statistically-based learning architectures: mixtures of Gaussians and locally weighted regression. While the techniques for neural networks are expensive and approximate, the techniques for mixtures of Gaussians and locally weighted regression are both efficient and accurate.

This thesis focuses on the numerical and theoretical studies of gas quenching in industrial furnaces. Gas quenching is the rapid cooling of metal pieces, aiming at forcing a phase transformation of the metal structure to improve its mechanical properties. The numerical methodology has been evaluated with respect to the desired accuracy and different aspects of the flow with importance for achieving an optimized process have been investigated. Initially, attention was paid to the flow and heat transfer fields both in an empty furnace and in a furnace loaded with different charges with the objective to study the influence of the charge configuration on the flow and heat transfer uniformity. This study led to the identification of several possible improvements, which are currently being implemented by the industrial partners of this project. As earlier studies had shown the importance of flow uniformity on the quality of the heat treatment, the subsequent work focused substantially on the flow uniformity upstream of the quenching zone resulting in design recommendations for the particular type of furnace under consideration. The dependence of the performance of the coolant medium on its composition was investigated theoretically and an analysis of most important parameters was carried out. Improved knowledge of the effect of gas mixture composition on heat transfer was added to the body of knowledge already available. / QC 20101019

This thesis develops models and associated Bayesian inference methods for flexible univariate and multivariate conditional density estimation. The models are flexible in the sense that they can capture widely differing shapes of the data. The estimation methods are specifically designed to achieve flexibility while still avoiding overfitting. The models are flexible both for a given covariate value, but also across covariate space. A key contribution of this thesis is that it provides general approaches of density estimation with highly efficient Markov chain Monte Carlo methods. The methods are illustrated on several challenging non-linear and non-normal datasets. In the first paper, a general model is proposed for flexibly estimating the density of a continuous response variable conditional on a possibly high-dimensional set of covariates. The model is a finite mixture of asymmetric student-t densities with covariate-dependent mixture weights. The four parameters of the components, the mean, degrees of freedom, scale and skewness, are all modeled as functions of the covariates. The second paper explores how well a smooth mixture of symmetric components can capture skewed data. Simulations and applications on real data show that including covariate-dependent skewness in the components can lead to substantially improved performance on skewed data, often using a much smaller number of components. We also introduce smooth mixtures of gamma and log-normal components to model positively-valued response variables. In the third paper we propose a multivariate Gaussian surface regression model that combines both additive splines and interactive splines, and a highly efficient MCMC algorithm that updates all the multi-dimensional knot locations jointly. We use shrinkage priors to avoid overfitting with different estimated shrinkage factors for the additive and surface part of the model, and also different shrinkage parameters for the different response variables. In the last paper we present a general Bayesian approach for directly modeling dependencies between variables as function of explanatory variables in a flexible copula context. In particular, the Joe-Clayton copula is extended to have covariate-dependent tail dependence and correlations. Posterior inference is carried out using a novel and efficient simulation method. The appendix of the thesis documents the computational implementation details. / <p>At the time of the doctoral defense, the following papers were unpublished and had a status as follows: Paper 3: In press. Paper 4: Manuscript.</p>

The overall objective of the work described in this thesis was to examine the effects of both waterborne and dietary routes of exposure to fathead minnow (Pimephales promelas) when exposed to complex metal mining mixtures. This was conducted using a 21-day, multi-trophic, short-term fathead minnow (FHM) reproductive bioassay. The endpoints that were measured were used to assess the effects on multiple levels of biological organization (sub-organismal to population endpoints). The first phase of this research was conducted in situ using environmentally realistic concentrations of 3 separate metal mining effluents [20% surface water effluent (SWE), 30% mine water effluent (MWE), 45% process water effluent (PWE)] from Sudbury, Ontario, Canada. Metals were analyzed in several media (water, sediments) and tissues (biofilm, Chironomus dilutus, female fathead minnow carcass, ovaries, liver and gills). The incorporation of the biofilm (primary producers) into the bioassay also added another level of organization that was novel to this study. Significant increases in metal concentrations were observed in the water and biofilm tissues in all treatments [SWE, MWE, PWE], compared to reference. Cobalt and nickel increased significantly in C. dilutus tissues in SWE (1.4-fold and 1.5-fold respectively), and copper and selenium in PWE (5.2-fold and 3.3-fold respectively), however no significant increases occurred in MWE compared to reference. There were no significant increases in metal concentrations in female FHM tissues (carcass, liver, gonads, gills) in any of the treatments, suggesting that metal bioavailability was reduced. Cumulative number of eggs per female per day increased significantly (+127%) after exposure to SWE and decreased significantly (-33%) after exposure to PWE when compared to the reference fish. Mean total number of days to hatch was also reduced in PWE compared to reference. In order to gain a better understanding of the routes of exposure causing toxicity in FHM, the second phase of this research examined the effects of exposure through diet, through water or through both using a fully factorial food exposure design in a laboratory setting. In this experiment we pre-exposed C. dilutus to both 45% PWE and laboratory control water until they reached the 3rd-4th instar stage of development (approximately 21 days) where they were collected and frozen until the start of the FHM reproductive bioassay. We further examined the role of food quality on fish toxicity by assessing differences between multi trophic (where fish were fed both a live and frozen diet of C. dilutus) in the laboratory. This research was conducted at the Toxicology Centre in Saskatoon, Saskatchewan, Canada. The results showed that significant effects were observed when fish were fed a live diet versus a frozen diet. Condition factor and body weight increased, although inconsistent effects were observed for liver somatic index (LSI) in fathead minnows in both experiments when exposed to one or both routes of exposure. Cumulative total egg production and cumulative spawning events were both significantly affected by both waterborne and dietborne exposures with the greatest effects seen in the multi-trophic streams and particularly when fish were fed a live diet. This significance of this research has demonstrated the importance of including both routes of exposure when assessing effects of mine effluent. This research also shows that the artificial stream technology is a useful tool in isolating the effects of a particular point source input (metal mining mixtures) when a system is highly confounded. The results suggest that under environmentally relevant exposure conditions, trophic transfer and live diet may lead to greater reproductive effects and increased fish toxicity. This also suggests that trophic transfer is an important route of exposure that is virtually impossible to attain using typical laboratory bioassay techniques (food-borne study using artificial diets or waterborne exposures only).

With the use of grand canonical molecular dynamics, we studied the slow ompression(0.01m/s) of very thin liquid films made of equimolar mixtures of short and long alkane chains (hexane and hexadecane), and branched and unbranched alkanes (phytane and hexadecane). Besides comparing how these mixtures behave under constant speed compression, we will compare their properties with the behavior and structure of the pure systems undergoing the same type of slow compression. To understand the arrangement of the molecules inside the confinement, we present segmental and molecular density profiles, average length and orientation of the molecules inside well layered gaps. To observe the effects of the compression on the fluids, we present the number of confined molecules, the inlayer orientation, the solvation force and the inlayer diffusion coefficient, versus the thickness of the gap. We observe that pure hexadecane, although liquid at this temperature, starts presenting strong solid-like behavior when it is compressed to thicknesses under 3nm, while pure hexane and pure phytane continue to behave liquid-like except at 1.3nm when they show some weak solid-like features. When hexadecane is mixed with the short straight hexane, it remains liquid down to 2.8nm at which point this mixture behaves solid-like with an enhanced alignment of the long molecules not seen in its pure form; but when hexadecane is mixed with the branched phytane the system does not present the solid-like features seen when hexadecane is compressed pure.

The learning process consists of different steps: building a Training Set (TS), training the system, testing its behaviour and finally classifying unknown objects. When using a distance based rule as a classifier, i.e. 1-Nearest Neighbour (1-NN), the first step (building a training set) includes editing and condensing data. The main reason for that is that the rules based on distance need many time to classify each unlabelled sample, x, as each distance from x to each point in the training set should be calculated. So, the more reduced the training set, the shorter the time needed for each new classification process. This thesis is mainly focused on building a training set from some already given data, and specially on condensing it; however different classification techniques are also compared.The aim of any condensing technique is to obtain a reduced training set in order to spend as few time as possible in classification. All that without a significant loss in classification accuracy. Somenew approaches to training set size reduction based on prototypes are presented. These schemes basically consist of defining a small number of prototypes that represent all the original instances. That includes those approaches that select among the already existing examples (selective condensing algorithms), and those which generate new representatives (adaptive condensing algorithms).Those new reduction techniques are experimentally compared to some traditional ones, for data represented in feature spaces. In order to test them, the classical 1-NN rule is here applied. However, other classifiers (fast classifiers) have been considered here, as linear and quadratic ones constructed in dissimilarity spaces based on prototypes, in order to realize how editing and condensing concepts work for this different family of classifiers.Although the goal of the algorithms proposed in this thesis is to obtain a strongly reduced set of representatives, the performance is empirically evaluated over eleven real data sets by comparing not only the reduction rate but also the classification accuracy with those of other condensing techniques. Therefore, the ultimate aim is not only to find a strongly reduced set, but also a balanced one.Several ways to solve the same problem could be found. So, in the case of using a rule based on distance as a classifier, not only the option of reducing the training set can be afford. A different family of approaches consists of applying several searching methods. Therefore, results obtained by the use of the algorithms here presented are compared in terms of classification accuracy and time, to several efficient search techniques.Finally, the main contributions of this PhD report could be briefly summarised in four principal points. Firstly, two selective algorithms based on the idea of surrounding neighbourhood. They obtain better results than other algorithms presented here, as well as better than other traditional schemes. Secondly, a generative approach based on mixtures of Gaussians. It presents better results in classification accuracy and size reduction than traditional adaptive algorithms, and similar to those of the LVQ. Thirdly, it is shown that classification rules other than the 1-NN can be used, even leading to better results. And finally, it is deduced from the experiments carried on, that with some databases (as the ones used here) the approaches here presented execute the classification processes in less time that the efficient search techniques.

No description available.

We give an upper bound for the norm distance of (0,1) -valued Markov-exchangeable random variables to mixtures of distributions of Markov processes. A Markov-exchangeable random variable has a distribution that depends only on the starting value and the number of transitions 0-0, 0-1, 1-0 and 1-1. We show that if, for increasing length of variables, the norm distance to mixtures of Markov processes goes to 0, the rate of this convergence may be arbitrarily slow. (author's abstract) / Series: Forschungsberichte / Institut für Statistik

The pendant drop technique was used to characterize the adsorption behavior of n-dodecane-1-thiol and n-hexane-1-thiol capped gold nanoparticles at the hexane-water interface. The adsorption process was studied by analyzing the dynamic interfacial tension versus nanoparticle concentration, both at early times and at later stages (i.e., immediately after the interface between the fluids is made and once equilibrium has been established). Following free diffusion of nanoparticles from the bulk hexane phase, adsorption leads to ordering and rearrangement of the nanoparticles at the interface and formation of a dense layer. With increasing interfacial coverage, the diffusion-controlled adsorption for the nanoparticles at the interface was found to change to an interaction-controlled assembly and the presence of an adsorption barrier was experimentally verified. At the same bulk concentration, different sizes of n-dodecane-1-thiol nanoparticles showed different absorption behavior at the interface, in agreement with the findings of Kutuzov et al. [1]. The experiments additionally demonstrated the important role played by the capping agent. At the same concentration, gold nanoparticles stabilized by n-hexane-1-thiol exhibited greater surface activity than gold nanoparticles of the same size stabilized by n-dodecane-1-thiol. 1.6 nm, 2.8 nm, and 4.4 nm nanoparticles capped with n-dodecane-1-thiol, and 2.9 nm, and 4.3 nm particles capped with n-hexane-1-thiol were used in this study. The physical size of the gold nanoparticles was determined by TEM image analysis. The pendant drop technique was also used to study the adsorption properties of mixtures of gold nanoparticles at the hexane-water interface; and also investigate the effects of different factors (i.e., temperature, pH or ionic strength) on interfacial tension (IFT). The interfacial properties of mixtures of these nanoparticles, having different sizes and capping agents, were then studied. No interaction was found between the unmixed studied nanoparticles. Using the theory of non-ideal interactions for binary mixtures, the interaction parameters for mixtures of nanoparticles at the interface were determined. The results indicate that nanoparticle concentration of the mixtures has a profound effect on the interfacial nanoparticle composition. A repulsive interaction between nanoparticles of different size and cap was found in the mixtures at the interface layer. The interfacial tension for mixtures was found to be higher than the interfacial tension for non-mixed nanoparticle suspensions. The nanoparticle composition at the interface was found to differ from the composition of nanoparticles in the bulk liquid phase. The activity of unmixed nanoparticles proved to be a poor predictor of the activity of mixtures. It was observed that the most active nanoparticles concentrated at the interface. The effects of temperature, pH and ionic strength concentration on the equilibrium and dynamic IFT of 4.4 nm gold nanoparticles capped with n-dodecane-1-thiol at the hydrocarbon-water interface was studied. The pendant drop technique was also used to study the adsorption properties of these nanoparticles at the hexane-water and nonane-water interface. The addition of NaCl was found to cause a decrease of the equilibrium and dynamic IFT greater than that, which accompanies the adsorption of nanoparticles at the interface in the absence of NaCl. Although IFT values for acidic and neutral conditions were found to be similar, a noticeable decrease in the IFT was found for more basic conditions. Increasing the temperature of the system was found to cause an increase in both dynamic and equilibrium IFT values. The adsorption of functionalized gold nanoparticles at liquid-liquid interfaces is a promising method for self-assembly and the creation of useful nanostructures. These findings contribute to the design of useful supra-colloidal structures by the self-assembly of alkane-thiol capped gold nanoparticles at liquid-liquid interfaces.

The overall objective of the work described in this thesis was to examine the effects of both waterborne and dietary routes of exposure to fathead minnow (Pimephales promelas) when exposed to complex metal mining mixtures. This was conducted using a 21-day, multi-trophic, short-term fathead minnow (FHM) reproductive bioassay. The endpoints that were measured were used to assess the effects on multiple levels of biological organization (sub-organismal to population endpoints). The first phase of this research was conducted in situ using environmentally realistic concentrations of 3 separate metal mining effluents [20% surface water effluent (SWE), 30% mine water effluent (MWE), 45% process water effluent (PWE)] from Sudbury, Ontario, Canada. Metals were analyzed in several media (water, sediments) and tissues (biofilm, Chironomus dilutus, female fathead minnow carcass, ovaries, liver and gills). The incorporation of the biofilm (primary producers) into the bioassay also added another level of organization that was novel to this study. Significant increases in metal concentrations were observed in the water and biofilm tissues in all treatments [SWE, MWE, PWE], compared to reference. Cobalt and nickel increased significantly in C. dilutus tissues in SWE (1.4-fold and 1.5-fold respectively), and copper and selenium in PWE (5.2-fold and 3.3-fold respectively), however no significant increases occurred in MWE compared to reference. There were no significant increases in metal concentrations in female FHM tissues (carcass, liver, gonads, gills) in any of the treatments, suggesting that metal bioavailability was reduced. Cumulative number of eggs per female per day increased significantly (+127%) after exposure to SWE and decreased significantly (-33%) after exposure to PWE when compared to the reference fish. Mean total number of days to hatch was also reduced in PWE compared to reference. In order to gain a better understanding of the routes of exposure causing toxicity in FHM, the second phase of this research examined the effects of exposure through diet, through water or through both using a fully factorial food exposure design in a laboratory setting. In this experiment we pre-exposed C. dilutus to both 45% PWE and laboratory control water until they reached the 3rd-4th instar stage of development (approximately 21 days) where they were collected and frozen until the start of the FHM reproductive bioassay. We further examined the role of food quality on fish toxicity by assessing differences between multi trophic (where fish were fed both a live and frozen diet of C. dilutus) in the laboratory. This research was conducted at the Toxicology Centre in Saskatoon, Saskatchewan, Canada. The results showed that significant effects were observed when fish were fed a live diet versus a frozen diet. Condition factor and body weight increased, although inconsistent effects were observed for liver somatic index (LSI) in fathead minnows in both experiments when exposed to one or both routes of exposure. Cumulative total egg production and cumulative spawning events were both significantly affected by both waterborne and dietborne exposures with the greatest effects seen in the multi-trophic streams and particularly when fish were fed a live diet. This significance of this research has demonstrated the importance of including both routes of exposure when assessing effects of mine effluent. This research also shows that the artificial stream technology is a useful tool in isolating the effects of a particular point source input (metal mining mixtures) when a system is highly confounded. The results suggest that under environmentally relevant exposure conditions, trophic transfer and live diet may lead to greater reproductive effects and increased fish toxicity. This also suggests that trophic transfer is an important route of exposure that is virtually impossible to attain using typical laboratory bioassay techniques (food-borne study using artificial diets or waterborne exposures only).