A new latent generalized dirichlet allocation model for image classification

As a response to the limitations of the LDA in topic modeling and large scale applications, several extensions using flexible priors have been introduced to expose the problem of topic correlation. Models such as CTM, PAM, GD-LDA, and LGDA have been able to explore and capture semantic relationships between topics. However, many of these models suffer from incomplete generative processes which affect inferences efficiency. In addition, knowing these traditional inference techniques carry major limitations, the new approach in this paper, the CVB-LGDA is an extension to the state-of-the-art. It reconciles a complete generative process to a robust inference technique in a topic correlation framework. Its performance in image classification shows its robustness.