KODAMA algorithm Came across this KODAMA algorithm that makes some pretty bold claims. I find the whole process hard to follow, as anyone looked at it/tried it? 
Isn't maximization of Cross-Validation accuracy "cheating/over fitting"? I find this a bit tricky, and was interested in knowing 1) about other hypothetical situations/algorithms where this criteria is applied and 2) if I am wrong in thinking this is just plain over fitting.
Link for the article
http://www.pnas.org/content/111/14/5117.abstract
 A: Here, I include the link a more recent publication on 
 Bioinformatics journal  and other information can be find on the vignette of the available R package.
Over-fitting is well-known issue in supervised learning. KODAMA is an unsupervised learning algorithm.
Over-fitting in unsupervised learning algorithms can also be an issue where the importance of sample noise is overestimated. An example is discussed by Balasubramanian and Schwart (Science 2002) relatively to the topological stability of the ISOMAP algorithm in presence of noisy data. Although, the parameter classifier selection is not in-depth discussed in the PNAS paper, it has demonstrated as KODAMA is more robust to the noise effect than the other effect.
In KODAMA output, the distance between two points gives us an indication of how much these two points can be predicted in the same classification group in a cross-validated model. Thus, this distance is a function of the classifier and its parameter.
Having said this, the results of KODAMA can be driven by the choose of the classifier and its relative parameter because the KODAMA dis
The result of KODAMA is mainly affected by the classification method and its relative parameter to use in the cross-validation procedure. This may be seen as disadvantage but it allow the user to see data from different point of view.
Yes, KODAMA cheats data but it doesn't over-fit them.
There are not other similar algorithm. KODAMA is the first of a new class of algorithm that use the maximization of the cross-validation accuracy to extract information from data.
