Abstract
In this paper, we propose a GPU acceleration of multi-topic extraction from images by using LDA (latent Dirichlet allocation). LDA is originally proposed as a probabilistic model for documents by Blei et al. In recent days, LDA is applied to multimedia information other than documents. We provide the results of experiments where we apply LDA to Professor Wang's 10,000 test images and extract multiple visional topics. We adpot collapsed variational Bayesian inference method for LDA and accelerate this by using Nvidia CUDA compatible GPU devices.
