Topic uncovering of the latent topics have become an active research area for more than a decade and continuous to receive contributions from all disciplines including computer science, information science and statistics. Since the introduction of Latent Dirichlet Allocation in 2003, many intriguing extension models have been proposed. One such extension model is the logistic normal correlated topic model, which not only uncovers hidden topic of a document, but also extract a meaningful topical relationship among a large number of topics. In this model, the Logistic normal distribution was adapted via the transformation of multivariate Gaussian variables to model the topical distribution of documents in the presence of correlations among topics. In this thesis, we propose a Probit normal alternative approach to modelling correlated topical structures. Our use of the Probit model in the context of topic discovery is novel, as many authors have so far concentrated solely of the logistic model partly due to the formidable inefficiency of the multinomial Probit model even in the case of very small topical spaces. We herein circumvent the inefficiency of multinomial Probit estimation by using an adaptation of the Diagonal Orthant Multinomial Probit (DO-Probit) in the topic models context, resulting in the ability of our topic modelling scheme to handle corpuses with a large number of latent topics. In addition, we extended our model and implement it into the context of image annotation by developing an efficient Collapsed Gibbs Sampling scheme. Furthermore, we employed various high performance computing techniques such as memory-aware Map Reduce, SpareseLDA implementation, vectorization and block sampling as well as some numerical efficiency strategy to allow fast and efficient sampling of our algorithm.
Library of Congress Subject Headings
Image analysis--Data processing; Probits; Multivariate analysis; Distribution (Probability theory)
Applied Statistics (MS)
Department, Program, or Center
The John D. Hromi Center for Quality and Applied Statistics (KGCOE)
Yu, Xingchen, "Topic Uncovering and Image Annotation via Scalable Probit Normal Correlated Topic Models" (2015). Thesis. Rochester Institute of Technology. Accessed from
RIT – Main Campus