Browsing by Subject "Hierarchical models"

Now showing 1 - 5 of 5

Bayesian learning methods for neural coding
(2013-12) Park, Mi Jung; Pillow, Jonathan W.; Bovik, Alan C. (Alan Conrad), 1958-
A primary goal in systems neuroscience is to understand how neural spike responses encode information about the external world. A popular approach to this problem is to build an explicit probabilistic model that characterizes the encoding relationship in terms of a cascade of stages: (1) linear dimensionality reduction of a high-dimensional stimulus space using a bank of filters or receptive fields (RFs); (2) a nonlinear function from filter outputs to spike rate; and (3) a stochastic spiking process with recurrent feedback. These models have described single- and multi-neuron spike responses in a wide variety of brain areas. This dissertation addresses Bayesian methods to efficiently estimate the linear and non-linear stages of the cascade encoding model. In the first part, the dissertation describes a novel Bayesian receptive field estimator based on a hierarchical prior that flexibly incorporates knowledge about the shapes of neural receptive fields. This estimator achieves error rates several times lower than existing methods, and can be applied to a variety of other neural inference problems such as extracting structure in fMRI data. The dissertation also presents active learning frameworks developed for receptive field estimation incorporating a hierarchical prior in real-time neurophysiology experiments. In addition, the dissertation describes a novel low-rank model for the high dimensional receptive field, combined with a hierarchical prior for more efficient receptive field estimation. In the second part, the dissertation describes new models for neural nonlinearities using Gaussian processes (GPs) and Bayesian active learning algorithms in closed-loop neurophysiology experiments to rapidly estimate neural nonlinearities. The dissertation also presents several stimulus selection criteria and compare their performance in neural nonlinearity estimation. Furthermore, the dissertation presents a variation of the new models by including an additional latent Gaussian noise source, to infer the degree of over-dispersion in neural spike responses. The proposed model successfully captures various mean-variance relationships in neural spike responses and achieves higher prediction accuracy than previous models.
Denizen politics : a comparative analysis of opposition to immigration in the European Union
(2014-08) Mohanty, Peter Cushner; Gregg, Benjamin Greenwood, 1954-; Givens, Terri E., 1964-; Luskin, Robert C.; Jessee, Stephen; Murer, Jeffrey
This dissertation presents a series of observational studies of opposition to immigration (OI) in the European Union. A substantial portion of the public seems to prefer a more exclusionary form of democracy, but how large, how vocal, and how organized that portion is varies considerably. I investigate exclusionism, a dimension of individual belief about how extensive political membership should be that tends to reflect how denizens prioritize political and cultural aspects of membership. In situating exclusionism, I shed light on three puzzles: Which of an individual’s concerns are the strongest determinants of OI? Which national developments are the strongest determinants of an individual’s OI? How are the effects of an individual’s concerns shaped by national context? Exclusionism predicts OI in more countries in the EU than do ideology or religion. Post-9/11 conflicts increase OI but not as dramatically as do increases in the Muslim population (suggesting perhaps that Islamophobia outpaces security risks). OI is highest in new countries of immigration, but polarization is most pronounced in older countries of immigration, where ongoing national developments have created unusually large generational gaps, religious differences, and disagreements about exclusionism. Political interest is key for explaining large differences in opinion, too. Exclusionism increases OI, even in low-immigration countries, among individuals with little interest in politics but only slightly; at high levels of individual interest and immigration, exclusionism’s effects are substantial. My findings reveal major challenges to integration policy in high-immigration countries: migrants and natives are unlikely to see eye-to-eye at any level of political interest, and there is near complete disagreement on immigration policy between politically-interested Muslims and politically-interested Christians. Methodologically, I introduce techniques to analyze polarization, and my findings have implications for best practices in cross-national survey research.
Document clustering with nonparametric hierarchical topic modeling
(2015-05) Schaefer, Kayla Hope; Williamson, Sinead; Zhou, Mingyuan
Since its introduction, topic modeling has been a fundamental tool in analyzing corpus structures. While the Relational Topic Model provides a way to link, and subsequently cluster, documents together as an extension of the original Latent Dirichlet Allocation (LDA) model, this paper seeks to form a document clustering model for the nonparametric alternative to LDA, the Dirichlet Process. As the structure of Shakespeare's tragedies is the focus of this work, we specifically cluster documents while modeling the text using a Hierarchical Dirichlet Process (HDP), which allows for a mixture model with shared mixture components, in order to capture the natural topic clustering within a play. Using collapsed Gibbs sampling, the effectiveness of the clustered HDP is compared against that of LDA and an HDP without document clustering. This is done using both log perplexity and a qualitative assessment of the returned topics. Furthermore, clustering is performed and analyzed individually on speeches from each of ten tragedies, as well as with a combined corpus of acts.
Ideology and interests : a hierarchical Bayesian approach to spatial party preferences
(2013-08) Mohanty, Peter Cushner; Jessee, Stephen A., 1980-
This paper presents a spatial utility model of support for multiple political parties. The model includes a "valence" term, which I reparameterize to include both party competence and the voters' key sociodemographic concerns. The paper shows how this spatial utility model can be interpreted as a hierarchical model using data from the 2009 European Elections Study. I estimate this model via Bayesian Markov Chain Monte Carlo (MCMC) using a block Gibbs sampler and show that the model can capture broad European-wide trends while allowing for significant amounts of heterogeneity. This approach, however, which assumes a normal dependent variable, is only able to partially reproduce the data generating process. I show that the data generating process can be reproduced more accurately with an ordered probit model. Finally, I discuss trade-offs between parsimony and descriptive richness and other practical challenges that may be encountered when v building models of party support and make recommendations for capturing the best of both approaches.
Modeling climate variables using Bayesian finite mixture models
(2015-05) Cuthbertson, Thomas Edwin; Keitt, Timothy H.; Müller, Peter
This paper presents an alternative to point-based clustering models using a Bayesian finite mixture model. Using a simulation of soil moisture data in the Amazon region of South America, a Bayesian mixture of regressions is used to preserve periodic behavior within clusters. The mixture model provides a full probabilistic description of all uncertainties in the parameters that generated the data in addition to a clustering algorithm which better preserves the periodic nature of data at a particular pixel.