Generating Data to Understand Data

It is about the research he has been doing during his sabbatical, it is very new and not even published. It takes a more abstract view on pattern mining and it uses algorithmic statistics to determine good sets of patterns (phrased differently: it also takes a more abstract view on MDL (the coding of the data is indirect)).
Many results that are only approximately true for Krimp (and only "proven'" experimentally for Krimp (because they are theoretically false)) are theoretically true for this new approach.

According to Arno 'it might be interesting only for die-hard pattern miners (and, probably, to philosophers of science, statisticians and other theoreticians)'.