Reproducibility page for Statistical Science paper

This is a review paper we were invited to write for a special issue on the EM algorithm for the journal of Statistical Science. The paper uses more mathematically explicit notation, and introduces the model and leraning by gradually developing the shared matching pursuit algorithm. The EM formulation is particularly useful for unsupervised learning. The webpage was last updated in October 2010.

Reproducibility page for ICCV paper

This is our first paper on this topic. The appendix shows that the linear additive model with Gaussian white noise residual is a special case of the density substitution scheme (which itself is a special case of maximum entropy tilting) for putting probability distribution on image intensities. The webpage was last updated in Feb 2008.

All the source code is copyrighted by the authors.