I know that the world doesn't really need another paper on a DEM method, but I've written one anyway.
I bring this psychological peculiarity to your attention only because along the way I've learned two very interesting things: [1] a Bayesian framework is useful for keeping a model with many degrees of freedom well behaved, and [2] there is a relatively new ensemble MCMC sampling technique that makes exploring high dimensional spaces possible. This method has few adjustable parameters and parallelizes well.
Best wishes,
Harry
Our paper is available on arXiv. It hasn't been submitted yet and comments are welcome. https://arxiv.org/abs/1610.05972
Some papers on the sampling method are available here: http://msp.org/camcos/2010/5-1/p04.xhtml http://adsabs.harvard.edu/abs/2013PASP..125..306F [795 citations!] http://adsabs.harvard.edu/abs/2013A%26C.....2...27A
You can find papers and slides on sparse Bayesian modeling here: http://www.miketipping.com/sparsebayes.htm
// // Harry P. Warren // Naval Research Laboratory // Code 7681 // Washington, DC 20375 //