The On-line Cross Entropy Method for Unsupervised Data Exploration

摘要

We investigate the use of the new Cross Entropy method as a tool for exploratory data analysis. We show how this method can be used to perform linear projections such as principal component analysis, exploratory projection pursuit and canonical correlation analysis. We further go on to show how topology preserving mappings can be created using the cross entropy method. We also show how the cross entropy method can be used to train deep architecture nets which are one of the main current research directions for creating true artificial intelligence. Finally we show how the cross entropy method can be used to optimize parameters for latent variable models.