The ability to organize concepts (e.g., dog, chair) into efficient mental representations, i.e., categories (e.g., animal, furniture) is a fundamental mechanism which allows humans to perceive, organize, and adapt to their world. Much research has been dedicated to the questions of how categories emerge and how they are represented. Experimental evidence suggests that (i) concepts and categories are represented through sets of features (e.g., dogs bark, chairs are made of wood) which are structured into different types (e.g, behavior, material); (ii) categories and their featural representations are learnt jointly and incrementally; and (iii) categories are dynamic and their representations adapt to changing environments. This thesis investigates the mechanisms underlying the incremental and dynamic formation of categories and their featural representations through cognitively motivated Bayesian computational models. Models of category acquisition have been extensively studied in cognitive science and primarily tested on perceptual abstractions or artificial stimuli. In this thesis, we focus on categories acquired from natural language stimuli, using nouns as a stand-in for their reference concepts, and their linguistic contexts as a representation of the concepts’ features. The use of text corpora allows us to (i) develop large-scale unsupervised models thus simulating human learning, and (ii) model child category acquisition, leveraging the linguistic input available to children in the form of transcribed child-directed language. In the first part of this thesis we investigate the incremental process of category acquisition. We present a Bayesian model and an incremental learning algorithm which sequentially integrates newly observed data. We evaluate our model output against gold standard categories (elicited experimentally from human participants), and show that high-quality categories are learnt both from child-directed data and from large, thematically unrestricted text corpora. We find that the model performs well even under constrained memory resources, resembling human cognitive limitations. While lists of representative features for categories emerge from this model, they are neither structured nor jointly optimized with the categories. We address these shortcomings in the second part of the thesis, and present a Bayesian model which jointly learns categories and structured featural representations. We present both batch and incremental learning algorithms, and demonstrate the model’s effectiveness on both encyclopedic and child-directed data. We show that high-quality categories and features emerge in the joint learning process, and that the structured features are intuitively interpretable through human plausibility judgment evaluation. In the third part of the thesis we turn to the dynamic nature of meaning: categories and their featural representations change over time, e.g., children distinguish some types of features (such as size and shade) less clearly than adults, and word meanings adapt to our ever changing environment and its structure. We present a dynamic Bayesian model of meaning change, which infers time-specific concept representations as a set of feature types and their prevalence, and captures their development as a smooth process. We analyze the development of concept representations in their complexity over time from child-directed data, and show that our model captures established patterns of child concept learning. We also apply our model to diachronic change of word meaning, modeling how word senses change internally and in prevalence over centuries. The contributions of this thesis are threefold. Firstly, we show that a variety of experimental results on the acquisition and representation of categories can be captured with computational models within the framework of Bayesian modeling. Secondly, we show that natural language text is an appropriate source of information for modeling categorization-related phenomena suggesting that the environmental structure that drives category formation is encoded in this data. Thirdly, we show that the experimental findings hold on a larger scale. Our models are trained and tested on a larger set of concepts and categories than is common in behavioral experiments and the categories and featural representations they can learn from linguistic text are in principle unrestricted.

Estudos recentes relacionados ao meio ambiente vêm ganhando grande destaque em todo o mundo devido ao fato dos níveis de poluição e a destruição das reservas naturais terem aumentado de maneira alarmante nos últimos anos. As grandes cidades são as que mais sofrem com a poluição e aqui serão estudados os níveis de poluição do ar em duas cidades em particular, a Cidade do México e São Paulo. A Cidade do México apresenta sérios problemas com os níveis de ozônio e São Paulo é a cidade brasileira com os maiores problemas relacionados à poluição. Entre os diferentes modelos considerados para analisar dados de poluição do ar, pode-se considerar o uso de modelos de séries temporais para modelar as médias diárias ou semanais de poluição. Nessa direção pode-se usar modelos de volatilidade estocástica. Essa família de modelos estatísticos tem sido extensivamente usada para analisar séries temporais financeiras, porém não se observa muitas aplicações em dados ambientais e de saúde. Modelos de volatilidade estocástica bivariados e multivariados, sob a aproximação Bayesiana, foram considerados para analisar os dados, especialmente usando métodos MCMC (Monte Carlo em Cadeias de Markov) para obter os sumários a posteriori de interesse, pois pode-se ter muitas dificuldades usando métodos clássicos de inferência estatística / Recent studies related to environmental has been considered in all world due to increasing levels of pollution and of natural resources destruction especially, in the last years. The largest cities in the world are the ones been mostly affected by pollution and in this work we consider the analysis of air pollution data of two important cities: Mexico City and São Paulo. The Mexico City presents serious problems of ozone levels and São Paulo is the Brazilian city with the largest problems related to air pollution. Among the different models which could be used to analyze air pollution data, we consider the use of time series modeling to the weekly or daily levels of pollution. In this way, we consider the use of volatility stochastic models. This family of models has been well explored with financial data but not well explored to analyze environmental and health data. Bivariate and multivariate stochastic models under the Bayesian approach were considered to analyze the data, especially using MCMC (Markov Chain Monte Carlo) methods to obtain the posterior summary of interest, since we usually have big difficulties using standard classical inference methods

This thesis develops a GIS-based Bayesian approach for area-wide traffic crash analysis. Five years of crash data from Houston, Texas, are analyzed using a geographic information system (GIS), and spatial-temporal patterns of relative crash risk are identified based on a hierarchical Bayesian approach. This Bayesian approach is used to filter the uncertainty in the data and identify and rank roadway segments with potentially high relative risks for crashes. The results provide a sound basis to take preventive actions to reduce the risks in these segments. To capture the real safety indications better, this thesis differentiates the risks in different directions of the roadways, disaggregates different road types, and utilizes GIS to analyze and visualize the spatial relative crash risks in 3-D views according to different temporal scales. Results demonstrate that the approach is effective in spatially smoothing the relative crash risks, eliminating the instability of estimates while maintaining real safety trends. The posterior risk maps show high-risk roadway segments in 3-D views, which is more reader friendly than the conventional 2-D views. The results are also useful for travelers to choose relatively safer routes.

<p>Coastal eutrophication is a complex process that is caused largely by anthropogenic nutrient enrichment. Estuaries are particularly susceptible to nutrient impairment, owing to their intimate connection with the contributing watersheds. Estuaries experiencing accelerating eutrophication are subject to a loss of key ecological functions and services. This doctoral dissertation presents the development and implementation of an integrated approach toward assessing the water quality in the Neuse Estuary following the implementation of the total maximum daily load (TMDL) program in the Neuse River basin. In order to accomplish this task, I have developed a series of water quality models and modeling strategies that can be effectively used in assessing nutrient based eutrophication. Two watershed-level nutrient loading models that operate on a different temporal scale are developed and used to quantify nitrogen loading to the Neuse Estuary over time. The models are used to probabilistically assess the success of the adopted mitigation measures in achieving the 30 % load reduction goal stipulated by the TMDL. Additionally, a novel structure learning approach is adopted to develop a Bayesian Network (BN) model that describes chlorophyll dynamics in the Upper Neuse Estuary. The developed BN model is compared to pre-TMDL models to assess any changes in the role that nutrient loading and physical forcings play in modulating chlorophyll levels in that section of the estuary. Finally, a set of empirical models are developed to assess the water quality monitoring program in the estuary, while also exploring the possibility of incorporating remotely sensed satellite data in an effort to augment the existing in-situ monitoring programs.</p> / Dissertation

<p>Under the increasing threat of climate change, it is imperative to understand the impact that environmental phenomena have on the demography and behavior of natural populations. In the last few decades an ever increasing body of research has documented dramatic changes in mortality rates and breeding phenology for a large number of species. A number of these have been attributed to the current trends in climate change, which have been particularly conspicuous in bird populations. However, datasets associated to these natural populations as well as to the environmental variables that affect their biology tend to be partial and incomplete. Thus, ecological research faces the urgent need to tackle these questions while at the same time develop inferential models that can handle the complex structure of these datasets and their associated uncertainty. Therefore, my dissertation research has focused on two main objectives: 1) to understand the relationship that demographic rates and breeding phenology of a colony of seabirds has with the environment in the context of climate change; and 2) to use and develop models that can encompass the complex structure of these natural systems, while also extending the process not only to inference but to building predictions. I divided this work in three research projects; for the first one I developed a hierarchical Bayesian model for age-specific survival for long lived species with capture-recapture data that allows the use of incomplete data (i.e. left-truncated and right-censored), and builds predictions of years of birth and death for all individuals while also drawing inference on the survivorship function. I compared this method to more traditional ones and address their limitations and advantages. My second research chapter makes use of this method to determine the age-specific survivorship of the Dry Tortugas sooty tern population, and explores the effect of changes in sea surface temperature on their cohort mortality rates. Finally, my third research chapter addresses the dramatic shift in breeding season experienced by the Dry Tortugas sooty tern colony, the most unprecedented shift reported for any bird species. I explore the role of climatic and weather variables as triggering mechanisms.</p> / Dissertation

This thesis develops a GIS-based Bayesian approach for area-wide traffic crash analysis. Five years of crash data from Houston, Texas, are analyzed using a geographic information system (GIS), and spatial-temporal patterns of relative crash risk are identified based on a hierarchical Bayesian approach. This Bayesian approach is used to filter the uncertainty in the data and identify and rank roadway segments with potentially high relative risks for crashes. The results provide a sound basis to take preventive actions to reduce the risks in these segments. To capture the real safety indications better, this thesis differentiates the risks in different directions of the roadways, disaggregates different road types, and utilizes GIS to analyze and visualize the spatial relative crash risks in 3-D views according to different temporal scales. Results demonstrate that the approach is effective in spatially smoothing the relative crash risks, eliminating the instability of estimates while maintaining real safety trends. The posterior risk maps show high-risk roadway segments in 3-D views, which is more reader friendly than the conventional 2-D views. The results are also useful for travelers to choose relatively safer routes.

Rare disease variants are receiving increasing importance in the past few years as the potential cause for many complex diseases, after the common disease variants failed to explain a large part of the missing heritability. With the advancement in sequencing techniques as well as computational capabilities, statistical methodology for analyzing rare variants is now a hot topic, especially in case-control association studies. In this thesis, we initially present two related statistical methodologies designed for case-control studies to predict the number of common and rare variants in a particular genomic region underlying the complex disease. Genome-wide association studies are nowadays routinely performed to identify a few putative marker loci or a candidate region for further analysis. These methods are designed to work with SNP data on such a genomic region highlighted by GWAS studies for potential disease variants. The fundamental idea is to use Bayesian methodology to obtain bivariate posterior distributions on counts of common and rare variants. While the first method uses randomly generated (minimal) ancestral recombination graphs, the second method uses ensemble clustering method to explore the space of genealogical trees that represent the inherent structure in the test subjects. In contrast to the aforesaid methods which work with SNP data, the third chapter deals with next-generation sequencing data to detect the presence of rare variants in a genomic region. We present a non-parametric statistical methodology for rare variant association testing, using the well-known Kolmogorov-Smirnov framework adapted for genetic data. it is a fast, model-free robust statistic, designed for situations where both deleterious and protective variants are present. It is also unique in utilizing the variant locations in the test statistic.

Made available in DSpace on 2015-03-26T13:32:20Z (GMT). No. of bitstreams: 1 texto completo.pdf: 475570 bytes, checksum: 32a4377514ec0978d86cb9bc9fcb45f1 (MD5) Previous issue date: 2013-07-31 / A hierarchical Bayesian modeling was used to evaluate the influence of halothane gene and its interaction with sex on pig &#769;s growth curves. Under this approach, the parameters from growth models (Logistic, Gompertz and von Bertalanffy) were estimated jointly with the effects of halothane gene and sex. A total of 344 F2 (Commercial x Piau) animals were weighted at birth, 21, 42, 63, 77, 105 and 150 days in life. The Logistic model has presented the best fit based on DIC (Deviance Information Criterion). Thus, the samples from marginal posterior distributions for the differences between the parameters estimates of Logistic model have indicated that the maturity weight of males with heterozygous genotypes (HALNn) was superior to males with homozygous genotypes (HALNN). In order to realize a comparison with the traditional methodology, the frequentist approach based on two distinct steps also was used, but there was not identified significant differences between growth curve parameter estimates from each group (combinations of halothane genotypes and sex). / Para avaliar a influência do gene halotano sobre a curva de crescimento de suínos, bem como sua interação com o sexo do animal, foi proposta uma modelagem hierárquica Bayesiana. Nesta abordagem, os parâmetros dos modelos não-lineares de crescimento (Logístico, Gompertz e von Bertalanffy) foram estimados conjuntamente com os efeitos de sexo e genótipos do gene halotano. Foram utilizados 344 animais F2(Comercial x Piau) pesados ao nascer, aos 21, 42, 63, 77, 105 e 150 dias. O modelo Logístico foi aquele que apresentou melhor qualidade de ajuste por apresentar menor DIC (Deviance Information Criterion) que os demais. As amostras das distribuições marginais a posteriori para as diferenças entre as estimativas dos parâmetros do modelo Logístico indicaram que o peso dos machos à idade adulta com genótipo heterozigoto (HALNn) foi superior ao dos homozigotos (HALNN). A título de comparação, também foi considerada a abordagem frequentista tradicional baseada em dois passos distintos, a qual, por apresentar um menor poder de discernimento estatístico, não mostrou diferenças significativas.

A travers les tâches de perception et de production, les humains peuvent manipuler non seulement des mots et des phrases mais également des unités de plus bas niveau tels des syllabes et des phonèmes. Les études en phonétique sont principalement focalisées sur ces seconds types d'unitées. Un des objectif majeur dans ce domaine et de comprendre comment les humains acquiert et manipulent ces unités.Dans cette thèse, nous nous intéressons à cette question à travers l'utilisation de la modélisation computationnelle en réalisant des simulation informatiques à l'aide d'un modèle bayésien de la communication nommé COSMO (“Communicating Objects using Sensory-Motor Operations”). Nos études s'étendres à trois aspects.Dans une première partie, nous étudions les représentations cognitives des unités phonétiques. Il est maintenant bien établie que les unités sont caractérisées par des représentations auditives et motrices. En examinant leur rôle respectifs durant le développement, nous établissons leur complémentarité à travers ce que nous nommons la propriété <<bande étroite/bande large>>.Dans une seconde partie, nous nous intéressons à la variabilité des unités phonétiques, notamment à travers l'étude de la corrélation des idiosyncrasies en perception et en production. En comparant plusieurs conditions de développement, nous établissons qu'elles s'acquiert à travers un processus de reproduction des catégories plutôt qu'à une répétition des sons.Dans une troisième partie, nous analysons la nature des catégories phonétiques. En phonétique, il y a un débat autour du statut des syllabes vs. des phonèmes dans la communication de la parole. Dans nos simulations, nous examinons leurs acquisitions respectives à travers un apprentissage non supervisée et montrons les particularités nécessaires à la communication. / Through perception and production tasks, humans are able to manipulate not only high-level units like words or sentences but also low-level units like syllables and phonemes. Studies in phonetics mainly focus on the second type of units. One of the main goal in this field is to understand how humans acquire and manipulate these units and how they are stored in the brain. In this PhD thesis, we address this set of issues by using computer modeling, performing computer simulations with a Bayesian model of communication, named COSMO (“Communicating Objects using Sensory-Motor Operations”). Our studies extend in three ways.In a first part, we investigate the cognitive content of phonetic units. It is well established that phonetic units are characterized by both auditory and motor representations. It also seems that these representations are both used during speech processing. We question the functional role of a double representation of phonetic units in the human brain, specifically in a perception task. By examining their respective development, we show that these two representations have a complementary role during perception: the auditory representation is tuned to recognize nominal stimuli whereas the motor representation has generalization properties and can deal with stimuli typical of adverse conditions. We call this the “auditory-narrow/motor-wide” property.In a second part, we investigate the variability of phonetic units. Despite the universality of phonetic units, their characterization varies from one person to another, both in their articulatory/motor and acoustic content. This is called idiosyncrasies. In our study, we aim at understanding how they appear during speech development. We specifically compare two learning algorithms, both based on an imitation process. The first version consists in sound imitation while the second version exploits phoneme imitation. We show that idiosyncrasies appear only in the course of a phoneme imitation process. We conclude that motor learning seems rather driven by a linguistic/communication goal than motivated by the reproduction of the stimulus acoustic properties.In a third part, we investigate the nature of phonetic units. In phonetics, there is a debate about the specific status of the syllable vs phoneme in speech communication. In adult studies, a consensus is now found: both units would be stored in the brain. But, in infant studies, syllabic units seem to be primary. In our simulation study, we investigate the acquisition of both units and try to understand how our model could “discover” phonemes starting from purely syllabic representations. We show that contrary to syllables and vowels, consonants are poorly characterized in the auditory representation, because the categories overlap. This is due to the influence of one phoneme on its neighbors, the well-known “coarticulation”. However, we also show that the representation of consonants in the motor space is much more efficient, with a very low level of overlap between categories. This is in line with classical theories about motor/articulatory invariance for plosives. In consequence, phonemes, i.e. vowels and consonants, seem well displayed and likely to clearly emerge in a sensory-motor developmental approach such as ours.Through these three axes, we implemented different versions of our model. Based on data from the literature, we specifically cared about the cognitive viability of its variables and distributions and of its learning phases. In this work, modeling computation has been used in two kinds of studies: comparative and explanatory studies. In the first ones, we compared results of two models differing by one aspect and we selected the one in accordance with experimental results. In the second ones, we interpreted a phenomenon observed in literature with our model. In both cases, our simulations aim at better understanding data from the literature and provide new predictions for future studies.

abstract: Hydrological models in arid and semi-arid ecosystems can be subject to high uncertainties. Spatial variability in soil moisture and evapotranspiration, key components of the water cycle, can contribute to model uncertainty. In particular, an understudied source of spatial variation is the effect of plant-plant interactions on water fluxes. At patch scales (plant and associated soil), plant neighbors can either negatively or positively affect soil water availability via competition or hydraulic redistribution, respectively. The aboveground microclimate can also be altered via canopy shading effects by neighbors. Across longer timescales (years), plants may adjust their physiological (water-use) traits in response to the neighbor-altered microclimate, which subsequently affects transpiration rates. The influence of physiological adjustments and neighbor-altered microclimate on water fluxes was assessed around Larrea tridentata in the Sonoran Desert. Field measurements of Larrea’s stomatal behavior and vertical root distributions were used to examine the effects of neighbors on Larrea’s physiological controls on transpiration. A modeling based approach was implemented to explore the sensitivity of evapotranspiration and soil moisture to neighbor effects. Neighbors significantly altered both above- and belowground physiological controls on evapotranspiration. Compared to Larrea growing alone, neighbors increased Larrea’s annual transpiration by up to 75% and 30% at the patch and stand scales, respectively. Estimates of annual transpiration were highly sensitive to the presence/absence of competition for water, and on seasonal timescales, physiological adjustments significantly influenced transpiration estimates. Plant-plant interactions can be a significant source of spatial variation in ecohydrological models, and both physiological adjustments to neighbors and neighbor effects on microclimate affect small scale (patch to ecosystem) water fluxes. / Dissertation/Thesis / Doctoral Dissertation Biology 2015