Methods for Hypothesis Testing in Animal Carcinogenicity Experiments

Thesis (Ph.D.)--University of Washington, 2016-08

Methods for Confounding Adjustment and High-Dimensional Environmental Exposures

Thesis (Ph.D.)--University of Washington, 2016-08

Convex Optimization Algorithms and Statistical Bounds for Learning Structured Models

Thesis (Ph.D.)--University of Washington, 2016-08

Multivariate statistical modeling of an anode backing furnace - Modélisation statistique multivariée du four à cuisson des anodes utilisées dans la fabrication d'aluminium primaire

La stratégie actuelle de contrôle de la qualité de l’anode est inadéquate pour détecter les anodes défectueuses avant qu’elles ne soient installées dans les cuves d’électrolyse. Des travaux antérieurs ont porté sur la modélisation du procédé de fabrication des anodes afin de prédire leurs propriétés directement après la cuisson en utilisant des méthodes statistiques multivariées. La stratégie de carottage des anodes utilisée à l’usine partenaire fait en sorte que ce modèle ne peut être utilisé que pour prédire les propriétés des anodes cuites aux positions les plus chaudes et les plus froides du four à cuire. Le travail actuel propose une stratégie pour considérer l’histoire thermique des anodes cuites à n’importe quelle position et permettre de prédire leurs propriétés. Il est montré qu’en combinant des variables binaires pour définir l’alvéole et la position de cuisson avec les données routinières mesurées sur le four à cuire, les profils de température des anodes cuites à différentes positions peuvent être prédits. Également, ces données ont été incluses dans le modèle pour la prédiction des propriétés des anodes. Les résultats de prédiction ont été validés en effectuant du carottage supplémentaire et les performances du modèle sont concluantes pour la densité apparente et réelle, la force de compression, la réactivité à l’air et le Lc et ce peu importe la position de cuisson.

The renormalization group via statistical inference

In physics, one attempts to infer the rules governing a system given only the results of imperfect measurements. Hence, microscopic theories may be effectively indistinguishable experimentally. We develop an operationally motivated procedure to identify the corresponding equivalence classes of states, and argue that the renormalization group (RG) arises from the inherent ambiguities associated with the classes: one encounters flow parameters as, e.g., a regulator, a scale, or a measure of precision, which specify representatives in a given equivalence class. This provides a unifying framework and reveals the role played by information in renormalization. We validate this idea by showing that it justifies the use of low-momenta n-point functions as statistically relevant observables around a Gaussian hypothesis. These results enable the calculation of distinguishability in quantum field theory. Our methods also provide a way to extend renormalization techniques to effective models which are not based on the usual quantum-field formalism, and elucidates the relationships between various type of RG.

The Effect of using online homework exercises on student achievement in a Quantitative methods course

Abstract: Quantitative Methods (QM) is a compulsory course in the Social Science program in CEGEP. Many QM instructors assign a number of homework exercises to give students the opportunity to practice the statistical methods, which enhances their learning. However, traditional written exercises have two significant disadvantages. The first is that the feedback process is often very slow. The second disadvantage is that written exercises can generate a large amount of correcting for the instructor. WeBWorK is an open-source system that allows instructors to write exercises which students answer online. Although originally designed to write exercises for math and science students, WeBWorK programming allows for the creation of a variety of questions which can be used in the Quantitative Methods course. Because many statistical exercises generate objective and quantitative answers, the system is able to instantly assess students’ responses and tell them whether they are right or wrong. This immediate feedback has been shown to be theoretically conducive to positive learning outcomes. In addition, the system can be set up to allow students to re-try the problem if they got it wrong. This has benefits both in terms of student motivation and reinforcing learning. Through the use of a quasi-experiment, this research project measured and analysed the effects of using WeBWorK exercises in the Quantitative Methods course at Vanier College. Three specific research questions were addressed. First, we looked at whether students who did the WeBWorK exercises got better grades than students who did written exercises. Second, we looked at whether students who completed more of the WeBWorK exercises got better grades than students who completed fewer of the WeBWorK exercises. Finally, we used a self-report survey to find out what students’ perceptions and opinions were of the WeBWorK and the written exercises. For the first research question, a crossover design was used in order to compare whether the group that did WeBWorK problems during one unit would score significantly higher on that unit test than the other group that did the written problems. We found no significant difference in grades between students who did the WeBWorK exercises and students who did the written exercises. The second research question looked at whether students who completed more of the WeBWorK exercises would get significantly higher grades than students who completed fewer of the WeBWorK exercises. The straight-line relationship between number of WeBWorK exercises completed and grades was positive in both groups. However, the correlation coefficients for these two variables showed no real pattern. Our third research question was investigated by using a survey to elicit students’ perceptions and opinions regarding the WeBWorK and written exercises. Students reported no difference in the amount of effort put into completing each type of exercise. Students were also asked to rate each type of exercise along six dimensions and a composite score was calculated. Overall, students gave a significantly higher score to the written exercises, and reported that they found the written exercises were better for understanding the basic statistical concepts and for learning the basic statistical methods. However, when presented with the choice of having only written or only WeBWorK exercises, slightly more students preferred or strongly preferred having only WeBWorK exercises. The results of this research suggest that the advantages of using WeBWorK to teach Quantitative Methods are variable. The WeBWorK system offers immediate feedback, which often seems to motivate students to try again if they do not have the correct answer. However, this does not necessarily translate into better performance on the written tests and on the final exam. What has been learned is that the WeBWorK system can be used by interested instructors to enhance student learning in the Quantitative Methods course. Further research may examine more specifically how this system can be used more effectively.

Mindfulness-Based Cognitive Therapy for older people in a community setting: a mixed methods feasibility study

Rationale: In line with complex intervention development, this research takes a systematic approach to examining the feasibility and acceptability of delivering Mindfulness-Based Cognitive Therapy (MBCT) to older people who experience symptoms of depression. Methods: A mixed methods approach was adopted in line with recommendations made by the MRC Complex Intervention Development framework. Quantitative and qualitative methods were combined by administering questionnaires as well as conducting post intervention interviews. A number of trial feasibility factors were examined such as recruitment and attrition rates. Qualitative data was analysed using Braun and Clarke’s thematic analysis framework. Results: Nine participants started the MBCT intervention and six completed the 8-week programme. The results suggest that MBCT for older people is feasible and acceptable. Participants reported improved mindfulness skills. Participants responded positively to being asked to take part in research and appeared to particularly value the group delivery format of the intervention. Conclusions: MBCT is both feasible and acceptable for older people experiencing symptoms of depression. Further research is required with larger sample sizes to allow for more robust statistical exploration of outcome measures, including mechanisms of change.

Statistical Analysis of Text Summarization Evaluation

This dissertation applies statistical methods to the evaluation of automatic summarization using data from the Text Analysis Conferences in 2008-2011. Several aspects of the evaluation framework itself are studied, including the statistical testing used to determine significant differences, the assessors, and the design of the experiment. In addition, a family of evaluation metrics is developed to predict the score an automatically generated summary would receive from a human judge and its results are demonstrated at the Text Analysis Conference. Finally, variations on the evaluation framework are studied and their relative merits considered. An over-arching theme of this dissertation is the application of standard statistical methods to data that does not conform to the usual testing assumptions.

Can we improve the statistical analysis of stroke trials? Statistical re-analysis of functional outcomes in stroke trials

Background: Most large acute stroke trials have been neutral. Functional outcome is usually analysed using a yes or no answer, e.g. death or dependency vs. independence. We assessed which statistical approaches are most efficient in analysing outcomes from stroke trials. Methods: Individual patient data from acute, rehabilitation and stroke unit trials studying the effects of interventions which alter functional outcome were assessed. Outcomes included modified Rankin Scale, Barthel Index, and ‘3 questions’. Data were analysed using a variety of approaches which compare two treatment groups. The results for each statistical test for each trial were then compared. Results: Data from 55 datasets were obtained (47 trials, 54,173 patients). The test results differed substantially so that approaches which use the ordered nature of functional outcome data (ordinal logistic regression, t-test, robust ranks test, bootstrapping the difference in mean rank) were more efficient statistically than those which collapse the data into 2 groups (chi square) (ANOVA p<0.001). The findings were consistent across different types and sizes of trial and for the different measures of functional outcome. Conclusions: When analysing functional outcome from stroke trials, statistical tests which use the original ordered data are more efficient and more likely to yield reliable results. Suitable approaches included ordinal logistic regression, t-test, and robust ranks test.

Statistical Analysis of the heavy neutral Atoms measured by IBEX

We investigate the directional distribution of heavy neutral atoms in the heliosphere by using heavy neutral maps generated with the IBEX-Lo instrument over three years from 2009 to 2011. The interstellar neutral (ISN) O&Ne gas flow was found in the first-year heavy neutral map at 601 keV and its flow direction and temperature were studied. However, due to the low counting statistics, researchers have not treated the full sky maps in detail. The main goal of this study is to evaluate the statistical significance of each pixel in the heavy neutral maps to get a better understanding of the directional distribution of heavy neutral atoms in the heliosphere. Here, we examine three statistical analysis methods: the signal-to-noise filter, the confidence limit method, and the cluster analysis method. These methods allow us to exclude background from areas where the heavy neutral signal is statistically significant. These methods also allow the consistent detection of heavy neutral atom structures. The main emission feature expands toward lower longitude and higher latitude from the observational peak of the ISN O&Ne gas flow. We call this emission the extended tail. It may be an imprint of the secondary oxygen atoms generated by charge exchange between ISN hydrogen atoms and oxygen ions in the outer heliosheath.

ArrayMining: a modular web-application for microarray analysis combining ensemble and consensus methods with cross-study normalization

Background: Statistical analysis of DNA microarray data provides a valuable diagnostic tool for the investigation of genetic components of diseases. To take advantage of the multitude of available data sets and analysis methods, it is desirable to combine both different algorithms and data from different studies. Applying ensemble learning, consensus clustering and cross-study normalization methods for this purpose in an almost fully automated process and linking different analysis modules together under a single interface would simplify many microarray analysis tasks. Results: We present ArrayMining.net, a web-application for microarray analysis that provides easy access to a wide choice of feature selection, clustering, prediction, gene set analysis and cross-study normalization methods. In contrast to other microarray-related web-tools, multiple algorithms and data sets for an analysis task can be combined using ensemble feature selection, ensemble prediction, consensus clustering and cross-platform data integration. By interlinking different analysis tools in a modular fashion, new exploratory routes become available, e.g. ensemble sample classification using features obtained from a gene set analysis and data from multiple studies. The analysis is further simplified by automatic parameter selection mechanisms and linkage to web tools and databases for functional annotation and literature mining. Conclusion: ArrayMining.net is a free web-application for microarray analysis combining a broad choice of algorithms based on ensemble and consensus methods, using automatic parameter selection and integration with annotation databases.

Probabilistic methods for seasonal forecasting in a changing climate: Cox-type regression models.

For climate risk management, cumulative distribution functions (CDFs) are an important source of information. They are ideally suited to compare probabilistic forecasts of primary (e.g. rainfall) or secondary data (e.g. crop yields). Summarised as CDFs, such forecasts allow an easy quantitative assessment of possible, alternative actions. Although the degree of uncertainty associated with CDF estimation could influence decisions, such information is rarely provided. Hence, we propose Cox-type regression models (CRMs) as a statistical framework for making inferences on CDFs in climate science. CRMs were designed for modelling probability distributions rather than just mean or median values. This makes the approach appealing for risk assessments where probabilities of extremes are often more informative than central tendency measures. CRMs are semi-parametric approaches originally designed for modelling risks arising from time-to-event data. Here we extend this original concept beyond time-dependent measures to other variables of interest. We also provide tools for estimating CDFs and surrounding uncertainty envelopes from empirical data. These statistical techniques intrinsically account for non-stationarities in time series that might be the result of climate change. This feature makes CRMs attractive candidates to investigate the feasibility of developing rigorous global circulation model (GCM)-CRM interfaces for provision of user-relevant forecasts. To demonstrate the applicability of CRMs, we present two examples for El Ni ? no/Southern Oscillation (ENSO)-based forecasts: the onset date of the wet season (Cairns, Australia) and total wet season rainfall (Quixeramobim, Brazil). This study emphasises the methodological aspects of CRMs rather than discussing merits or limitations of the ENSO-based predictors.

Statistical learning of random probability measures

The study of random probability measures is a lively research topic that has attracted interest from different fields in recent years. In this thesis, we consider random probability measures in the context of Bayesian nonparametrics, where the law of a random probability measure is used as prior distribution, and in the context of distributional data analysis, where the goal is to perform inference given avsample from the law of a random probability measure. The contributions contained in this thesis can be subdivided according to three different topics: (i) the use of almost surely discrete repulsive random measures (i.e., whose support points are well separated) for Bayesian model-based clustering, (ii) the proposal of new laws for collections of random probability measures for Bayesian density estimation of partially exchangeable data subdivided into different groups, and (iii) the study of principal component analysis and regression models for probability distributions seen as elements of the 2-Wasserstein space. Specifically, for point (i) above we propose an efficient Markov chain Monte Carlo algorithm for posterior inference, which sidesteps the need of split-merge reversible jump moves typically associated with poor performance, we propose a model for clustering high-dimensional data by introducing a novel class of anisotropic determinantal point processes, and study the distributional properties of the repulsive measures, shedding light on important theoretical results which enable more principled prior elicitation and more efficient posterior simulation algorithms. For point (ii) above, we consider several models suitable for clustering homogeneous populations, inducing spatial dependence across groups of data, extracting the characteristic traits common to all the data-groups, and propose a novel vector autoregressive model to study of growth curves of Singaporean kids. Finally, for point (iii), we propose a novel class of projected statistical methods for distributional data analysis for measures on the real line and on the unit-circle.

Concepts and methods for efficient decentralized learning in federated settings

Deep Neural Networks (DNNs) have revolutionized a wide range of applications beyond traditional machine learning and artificial intelligence fields, e.g., computer vision, healthcare, natural language processing and others. At the same time, edge devices have become central in our society, generating an unprecedented amount of data which could be used to train data-hungry models such as DNNs. However, the potentially sensitive or confidential nature of gathered data poses privacy concerns when storing and processing them in centralized locations. To this purpose, decentralized learning decouples model training from the need of directly accessing raw data, by alternating on-device training and periodic communications. The ability of distilling knowledge from decentralized data, however, comes at the cost of facing more challenging learning settings, such as coping with heterogeneous hardware and network connectivity, statistical diversity of data, and ensuring verifiable privacy guarantees. This Thesis proposes an extensive overview of decentralized learning literature, including a novel taxonomy and a detailed description of the most relevant system-level contributions in the related literature for privacy, communication efficiency, data and system heterogeneity, and poisoning defense. Next, this Thesis presents the design of an original solution to tackle communication efficiency and system heterogeneity, and empirically evaluates it on federated settings. For communication efficiency, an original method, specifically designed for Convolutional Neural Networks, is also described and evaluated against the state-of-the-art. Furthermore, this Thesis provides an in-depth review of recently proposed methods to tackle the performance degradation introduced by data heterogeneity, followed by empirical evaluations on challenging data distributions, highlighting strengths and possible weaknesses of the considered solutions. Finally, this Thesis presents a novel perspective on the usage of Knowledge Distillation as a mean for optimizing decentralized learning systems in settings characterized by data heterogeneity or system heterogeneity. Our vision on relevant future research directions close the manuscript.