Attentional set-shifting tasks, consisting of multiple stages of discrimination learning, have been widely used in animals and humans to investigate behavioural flexibility. However, there are several learning criteria (e.g., 6-correct-choice-in-a-row, or 10-out- of-12-correct) by which a subject might be judged to have learned a discrimination. Furthermore, the currently frequentist approach does not provide a detailed analysis of individual performance. In this PhD study, a large set of archival data of rats performing a 7-stage intra-dimensional/extra-dimensional (ID/ED) attentional set- shifting task was analysed, using a novel Bayesian analytical approach, to estimate each rat's learning processes over its trials within the task. The analysis showed that the Bayesian learning criterion may be an appropriate alternative to the frequentist n- correct-in-a-row criterion for studying performance. The individual analysis of rats' behaviour using the Bayesian model also suggested that the rats responded according to a number of irrelevant spatial and perceptual information sources before the correct stimulus-reward association was established. The efficacy of the Bayesian analysis of individual subjects' behaviour and the appropriateness of the Bayesian learning criterion were also supported by the analysis of simulated data in which the behavioural choices in the task were generated by known rules. Additionally, the efficacy was also supported by analysis of human behaviour during an analogous human 7-stage attentional set-shifting task, where participants' detailed learning processes were collected based on their trial-by-trial oral report. Further, an extended Bayesian approach, which considers the effects of feedback (correct vs incorrect) after each response in the task, can even help infer whether individual human participants have formed an attentional set, which is crucial when applying the set-shifting task to an evaluation of cognitive flexibility. Overall, this study demonstrates that the Bayesian approach can yield additional information not available to the conventional frequentist approach. Future work could include refining the rat Bayesian model and the development of an adaptive trial design.

Orientador: Jacques Wainer / Tese (doutorado) - Universidade Estadual de Campinas, Instituto de Computação / Made available in DSpace on 2018-08-09T21:11:41Z (GMT). No. of bitstreams: 1 JambeiroFilho_JorgeEduardodeSchoucair_D.pdf: 736285 bytes, checksum: b7d7f186f743f9b0e541c857b0ca8226 (MD5) Previous issue date: 2007 / Resumo: Analisamos o uso de métodos Bayesianos em um problema de classificação de padrões de interesse prático para a Receita Federal do Brasil que é caracterizado pela presença de atributos de alta cardinalidade e pela existência de interações relevantes entre eles. Mostramos que a presença de atributos de alta cardinalidade pode facilmente gerar tantas subdivisões no conjunto de treinamento que, mesmo tendo originalmente uma grande quantidade de dados, acabemos obtendo probabilidades pouco confiáveis, inferidas a partir de poucos exemplos. Revisamos as estratégias usualmente adotadas para lidar com esse problema dentro do universo Bayesiano, exibindo sua dependência em suposições de não interação inaceitáveis em nosso domínio alvo. Mostramos empiricamente que estratégias Bayesianas mais avançadas para tratamento de atributos de alta cardinalidade, como pré-processamento para redução de cardinalidade e substituição de tabelas de probabilidades condicionais (CPTs) de redes Bayesianas (BNs) por tabelas default (DFs), árvores de decisão (DTs) e grafos de decisão (DGs) embora tragam benefícios pontuais não resultam em ganho de desempenho geral em nosso domínio alvo. Propomos um novo método Bayesiano de classificação, chamado de hierarchical pattern Bayes (HPB), que calcula probabilidades posteriores para as classes dado um padrão W combinando as observações de W no conjunto de treinamento com probabilidades prévias que são obtidas recursivamente a partir das observações de padrões estritamente mais genéricos que W. Com esta estratégia, ele consegue capturar interações entre atributos de alta cardinalidade quando há dados suficientes para tal, sem gerar probabilidades pouco confiáveis quando isso não ocorre. Mostramos empiricamente que, em nosso domínio alvo, o HPB traz benefícios significativos com relação a redes Bayesianas com estruturas populares como o naïve Bayes e o tree augmented naïve Bayes, com relação a redes Bayesianas (BNs) onde as tabelas de probabilidades condicionais foram substituídas pelo noisy-OR, por DFs, por DTs e por DGs, e com relação a BNs construídas, após uma fase de redução de cardinalidade usando o agglomerative information bottleneck. Além disso, explicamos como o HPB, pode substituir CPTs e mostramos com testes em outro problema de interesse prático que esta substituição pode trazer ganhos significativos. Por fim, com testes em vários conjuntos de dados públicos da UCI, mostramos que a utilidade do HPB ser bastante ampla / Abstract: In this work, we analyze the use of Bayesian methods in a pattern classification problem of practical interest for Brazil¿s Federal Revenue which is characterized by the presence of high cardinality attributes and by the existence of relevant interactions among them.We show that the presence of high cardinality attributes can easily produce so many subdivisions in the training set that, even having originally a great amount of data, we end up with unreliable probability estimates, inferred from small samples. We cover the most common strategies to deal with this problem within the Bayesian universe and show that they rely strongly on non interaction assumptions that are unacceptable in our target domain. We show empirically that more advanced strategies to handle high cardinality attributes like cardinality reduction by preprocessing and conditional probability tables replacement with default tables, decision trees and decision graphs, in spite of some restricted benefits, do not improve overall performance in our target domain. We propose a new Bayesian classification method, named hierarchical pattern Bayes (HPB), which calculates posterior class probabilities given a pattern W combining the observations of W in the training set with prior class probabilities that are obtained recursively from the observations of patterns that are strictly more generic than W. This way, it can capture interactions among high cardinality attributes when there is enough data, without producing unreliable probabilities when there is not. We show empirically that, in our target domain, HPB achieves significant performance improvements over Bayesian networks with popular structures like naïve Bayes and tree augmented naïve Bayes, over Bayesian networks where traditional conditional probability tables were substituted by noisy-OR gates, default tables, decision trees and decision graphs, and over Bayesian networks constructed after a cardinality reduction preprocessing phase using the agglomerative information bottleneck method. Moreover, we explain how HPB can replace conditional probability tables of Bayesian Networks and show, with tests in another practical problem, that such replacement can result in significant benefits. At last, with tests over several UCI datasets we show that HPB may have a quite wide applicability / Doutorado / Sistemas de Informação / Doutor em Ciência da Computação

Social learning is ubiquitous among animals and humans and is thought to be critical to the widespread success of humans and to the development and evolution of human culture. Evolutionary theory, however, suggests that social learning alone may not be adaptive but that individuals may need to be selective in who and how they copy others. One of the key findings of these evolutionary models (reviewed in Chapter 1) is that social information may be widely adaptive if individuals are able to combine social and asocial sources of information together strategically. However, up until this point the focus of theoretic models has been on the population level consequences of different social learning strategies, and not on how individuals combine social and asocial information on specific tasks. In Chapter 2 I carry out an analysis of how animal learners might incorporate social information into a reinforcement learning framework and find that even limited, low-fidelity copying of actions in an action sequence may combine with asocial learning to result in high fidelity transmission of entire action sequences. In Chapter 3 I describe a series of experiments that find that human learners flexibly use a conformity biased learning strategy to learn from multiple demonstrators depending on demonstrator accuracy, either indicated by environmental cues or past experience with these demonstrators. The chapter reveals close quantitative and qualitative matches between participant's performance and a Bayesian model of social learning. In both Chapters 2 and 3 I find, consistent with previous evolutionary findings, that by combining social and asocial sources of information together individuals are able to learn about the world effectively. Exploring how animals use social learning experimentally can be a substantially more difficult task than exploring human social learning. In Chapter 4, I develop and present a refined version of Network Based Diffusion analysis to provide a statistical framework for inferring social learning mechanisms from animal diffusion experiments. In Chapter 5 I move from examining the effects of social learning at an individual level to examining their population level outcomes and provide an analysis of how fine-grained population structure may alter the spread of novel behaviours through a population. I find that although a learner's social learning strategy and the learnability of a novel behaviour strongly impact how likely the behaviour is to spread through the population, fine grained population structure plays a much smaller role. In Chapter 6 I summarize the results of this thesis, and provide suggestions for future work to understand how individuals, humans and other animals alike, use social information.

Disease mapping and spatial statistics have become an important part of modern day statistics and have increased in popularity as the methods and techniques have evolved. The application of disease mapping is not only confined to the analysis of diseases as other applications of disease mapping can be found in Econometric and financial disciplines. This thesis will consider two data sets. These are the Georgia oral cancer 2004 data set and the South African acute pericarditis 2014 data set. The Georgia data set will be used to assess the hyperprior sensitivity of the precision for the uncorrelated heterogeneity and correlated heterogeneity components in a convolution model. The correlated heterogeneity will be modelled by a conditional autoregressive prior distribution and the uncorrelated heterogeneity will be modelled with a zero mean Gaussian prior distribution. The sensitivity analysis will be performed using three models with conjugate, Jeffreys' and a fixed parameter prior for the hyperprior distribution of the precision for the uncorrelated heterogeneity component. A simulation study will be done to compare four prior distributions which will be the conjugate, Jeffreys', probability matching and divergence priors. The three models will be fitted in WinBUGS® using a Bayesian approach. The results of the three models will be in the form of disease maps, figures and tables. The results show that the hyperprior of the precision for the uncorrelated heterogeneity and correlated heterogeneity components are sensitive to changes and will result in different results depending on the specification of the hyperprior distribution of the precision for the two components in the model. The South African data set will be used to examine whether there is a difference between the proper conditional autoregressive prior and intrinsic conditional autoregressive prior for the correlated heterogeneity component in a convolution model. Two models will be fitted in WinBUGS® for this comparison. Both the hyperpriors of the precision for the uncorrelated heterogeneity and correlated heterogeneity components will be modelled using a Jeffreys' prior distribution. The results show that there is no significant difference between the results of the model with a proper conditional autoregressive prior and intrinsic conditional autoregressive prior for the South African data, although there are a few disadvantages of using a proper conditional autoregressive prior for the correlated heterogeneity which will be stated in the conclusion.

Decision makers have long relied on experts to inform their decision making. Expert judgment analysis is a way to elicit and combine the opinions of a group of experts to facilitate decision making. The use of expert judgment is most appropriate when there is a lack of data for obtaining reasonable statistical results. The experts are asked for advice by one or more decision makers who face a specific real decision problem. The decision makers are outside the group of experts and are jointly responsible and accountable for the decision and committed to finding solutions that everyone can live with. The emphasis is on the decision makers learning from the experts. The focus of this thesis is an overview and comparison of the various elicitation and combination methods available. These include the traditional committee method, the Delphi method, the paired comparisons method, the negative exponential model, Cooke’s classical model, the histogram technique, using the Dirichlet distribution in the case of a set of uncertain proportions which must sum to one, and the employment of overfitting. The supra Bayes approach, the determination of weights for the experts, and combining the opinions of experts where each opinion is associated with a confidence level that represents the expert’s conviction of his own judgment are also considered.

This thesis is concerned with the problem of how to make decisions in an uncertain world. We use a model of uncertainty based on Markov decision problems, and develop a number of algorithms for decision-making both for the planning problem, in which the model is known in advance, and for the reinforcement learning problem in which the decision-making agent does not know the model and must learn to make good decisions by trial and error. The basis for much of this work is the use of structural representations of problems. If a problem is represented in a structured way we can compute or learn plans that take advantage of this structure for computational gains. This is because the structure allows us to perform abstraction. Rather than reasoning about each situation in which a decision must be made individually, abstraction allows us to group situations together and reason about a whole set of them in a single step. Our approach to abstraction has the additional advantage that we can dynamically change the level of abstraction, splitting a group of situations in two if they need to be reasoned about separately to find an acceptable plan, or merging two groups together if they no longer need to be distinguished. We present two planning algorithms and one learning algorithm that use this approach. A second idea we present in this thesis is a novel approach to the exploration problem in reinforcement learning. The problem is to select actions to perform given that we would like good performance now and in the future. We can select the current best action to perform, but this may prevent us from discovering that another action is better, or we can take an exploratory action, but we risk performing poorly now as a result. Our Bayesian approach makes this tradeoff explicit by representing our uncertainty about the values of states and using this measure of uncertainty to estimate the value of the information we could gain by performing each action. We present both model-free and model-based reinforcement learning algorithms that make use of this exploration technique. Finally, we show how these ideas fit together to produce a reinforcement learning algorithm that uses structure to represent both the problem being solved and the plan it learns, and that selects actions to perform in order to learn using our Bayesian approach to exploration. / Science, Faculty of / Computer Science, Department of / Graduate

This thesis presents a methodology for obtaining the optimal design capacity for sediment yield in multipurpose reservoir design. A stochastic model is presented for the prediction of sediment yield in a semi -arid watershed based on rainfall data and watershed characteristics. Uncertainty stems from each of the random variables used in the model, namely, rainfall amount, storm duration, runoff, peak flow rate, and number of events per season. Using the stochastic sediment yield model for N- seasons, a Bayesian decision analysis is carried out for a dam site in southern Arizona. Extensive numerical analyses and simplifying assumptions are made to facilitate finding the optimal solution. The model has applications in the planning of reservoirs and dams where the effective lifetime of the facility may be evaluated in terms of storage capacity and of the effects of land management on the watershed. Experimental data from the Atterbury watershed are used to calibrate the model and to evaluate uncertainties associated with our knowledge of the parameters of the joint distribution of rainfall and storm duration used in calculating the sediment yield amount.

Orientadores: Clarice Weis Arns, Márcia Bianchi dos Santos / Tese (doutorado) - Universidade Estadual de Campinas, Instituto de Biologia / Made available in DSpace on 2018-08-27T11:13:27Z (GMT). No. of bitstreams: 1 Carvalho_RicardoDuraesde_D.pdf: 3518273 bytes, checksum: 7b6f8b159eb057429823e23f6852c29b (MD5) Previous issue date: 2015 / Resumo: A evolução e a dinâmica populacional dos Coronavírus (CoVs) ainda permanecem pouco exploradas. No presente estudo, análises filogenéticas e de filogeografia foram conduzidas para investigar a dinâmica evolutiva dos CoVs detectados em aves silvestres e sinantrópicas. Um total de 500 amostras, que inclui os suabes traqueais e cloacais coletados de 312 aves silvestres pertencentes a 42 espécies, foram analisadas através da RT-qPCR. Sessenta e cinco amostras (13%) provenientes de 23 espécies foram positivas para o Coronavírus aviário (AvCoV). Trezentos e duas amostras foram investigadas para a pesquisa do Pan-Coronavírus (Pan-CoV) através do nPCR, destas, 17 (5,6%) foram positivas, sendo que 11 foram detectadas em espécies diferentes. Análises filogenéticas dos AvCoVs revelaram que as sequências de DNA das amostras coletadas no Brasil não agruparam com nenhuma das sequências do gene Spike (S1) dos AvCoVs depositados no banco de dados GenBank. Análise Bayesiana estimou uma variante do AvCoV proveniente da Suécia (1999) como o ancestral comum mais recente dos AvCoVs detectados neste estudo. Além disso, as análises realizadas através do "Bayesian Skyline Plot" (BSP) inferiram um aumento na dinâmica da população demográfica do AvCoV em diferentes espécies de aves silvestres e sinantrópicas. As análises filogenéticas do Pan-CoV mostrou que a maioria das amostras se agruparam com o Vírus da Hepatite Murina A59 (MHV A59), CoV pertencente ao grupo dos Beta-CoVs. Uma amostra [CoV detectado em Amazona vinacea(Papagaio-de-peito-roxo)] se agrupou com um CoV de Suínos, o PCoV HKU15, que pertence ao gênero Delta-CoV, ainda não relatado na América do Sul. Nossos achados sugerem que as aves podem ser novos potenciais hospedeiros responsáveis pela propagação e disseminação de diferentes CoVs para diferentes espécies de animais / Abstract: The evolution and population dynamics of Coronaviruses (CoVs) still remain underexplored. In the present study, phylogenetic and phylogeographic analyseswere conducted to investigate the evolutionary dynamics of CoV detected in wild and synanthropic birds. A total of 500 samples, including tracheal and cloacal swabs collected from 312 wild birds belonging to 42species, were analysed by RT-qPCR. A total of 65 samples from 23bird species were positive for Avian Coronaviruses (AvCoVs).Three hundred and two samples were screened for the Pan-Coronavirus (Pan-CoV) through the nPCR, 17 (5.6%) were positive, being that 11 were detected in different species. AvCoVs phylogenetic analyses revealed that the DNA sequences from samples collected in Brazil did not cluster with any of the AvCoV S1 gene sequences deposited in the GenBank database. Bayesian framework analysis estimated an AvCoV strain from Sweden (1999) as the most recent common ancestor of the AvCoVs detected in this study. Furthermore, Bayesian Skyline Plot (BSP) analysis inferred an increase in the AvCoV dynamic demographic population in different wild and synanthropic bird species. Phylogenetic analysis of the Pan-CoV showed that most of the samples clustered with the Murine Hepatitis Virus A59 strain (MHV A59) belong to the BetaCoV group. Besides, one of our samples [CoV detected in Amazona vinacea (parrot-breasted-purple)] clustered with a CoV isolated from pigs, PCoV HKU15, belonging to the DeltaCoV genus, still not reported in South America. Our findings suggest that birds may be potential new hosts responsible for spreading of different CoVs for different species of animals / Doutorado / Microbiologia / Doutor em Genetica e Biologia Molecular

This thesis explores the use of unstructured data, and specifically textual data, in providing consumer insights and improving business decisions. The thesis consists of two essays. In essay I, I examine how the linguistic similarity between the language used by reviewers of a product and a prospective customer’s own writing style can be leveraged to assess the match between customers and products. Applying tools from machine learning, Bayesian statistics, and computational linguistics to a large-scale dataset from Yelp, I find that the closer the writing style of a restaurant’s past reviews are to a prospective customer’s writing style, the more likely that customer is to write a review for that restaurant. This effect holds across restaurant types and is driven by the linguistic similarity between the customer’s own reviews and positive past reviews for the restaurant. Further, I find that similarity with respect to words related to leisure (e.g., family, wine, beer, weekend), biology (e.g., eat, life, love), as well as swear words are most influential in creating a match between customers and restaurants. In essay II, I examine whether borrowers consciously or not, leave traces of their intentions, circumstances, and personality traits in the text they write when applying for a loan. I find that this textual information has a substantial and significant ability to predict whether borrowers will pay back the loan above and beyond the financial and demographic variables commonly used in models predicting default. Using text-mining and machine-learning tools to automatically process and analyze the raw text in over 120 thousand loan requests from Prosper.com, an online crowdfunding platform, I find that including the textual information in the loan significantly helps predict loan default and can have substantial financial implications. I find that loan requests written by defaulting borrowers are more likely to include words related to their family, mentions of God, the borrower’s financial and general hardship, pleading lenders for help, and short-term focused words. I further observe that defaulting loan requests are written in a manner consistent with the writing style of extroverts and liars.

This thesis studies event history analysis in multivariate longitudinal observational databases (LODs) and its application in postmarketing surveillance to identify and measure the relationship between events of health outcomes and drug exposures. The LODs contain repeated measurements on each individual whose healthcare information is recorded electronically. Novel statistical methods are being developed to handle challenging issues arising from the scale and complexity of postmarketing surveillance LODs. In particular, the self-controlled case series (SCCS) method has been developed with two major features (1) it only uses individuals with at least one event for analysis and inference and, (2) it uses each individual to be served as his/her own control, effectively requiring a person to switch treatments during the observation period. Although this method handles heterogeneity and bias, it does not take full advantage of the observational databases. In this connection, the SCCS method may lead to a substantial loss of efficiency. We proposed a multivariate proportional intensity modeling approach with random effect for multivariate LODs. The proposed method can explain the heterogeneity and eliminate bias in LODs. It also handles multiple types of event cases and makes full use of the observational databases. In the first part of this thesis, we present the multivariate proportional intensity model with correlated frailty. We explore the correlation structure between multiple types of clinical events and drug exposures. We introduce a multivariate Gaussian frailty to incorporate thewithin-subject heterogeneity, i.e. hidden confounding factors. For parameter estimation, we adopt the Bayesian approach using the Markov chain Monte Carlo method to get a series of samples from the targeted full likelihood. We compare the new method with the SCCS method and some frailty models through simulation studies. We apply the proposed model to an electronic health record (EHR) dataset and identify event types as defined in Observational Outcomes Medical Partnership (OMOP) project. We show that the proposed method outperforms the existing methods in terms of common metrics, such as receiver operating characteristic (ROC) metrics. Finally, we extend the proposed correlated frailty model to include a dynamic random effect. We establish a general asymptotic theory for the nonparametric maximum likelihood estimators in terms of identifiability, consistency, asymptotic normality and asymptotic efficiency. A detailed illustration of the proposed method is done with the clinical event Myocardial Infarction (MI) and drug treatment of Angiotensin-converting-enzyme (ACE) inhibitors, showing the dynamic effect of unobserved heterogeneity.