The past six weeks on vacation and at the Milano WCC conference had me sort my ideas into a likely dissertation structure. It is somewhat a follow-up on this AI-2008 announcement article, although the ideas in that blog post are not exactly what ended up being in the conference paper (self-organizing maps).

The dissertation structure might look as follows:

  • Description and definition of data and scope of the data mining process
  • Description and definition of models (theoretical work)
  • Use the modeling techniques on the data
  • Find new heterogeneity indicators (or evaluate existing ones)

From the ideas and the publications so far, the NN/MLP stuff would fit nicely as a reference model to compare other models with. The NN/SOM stuff probably fits best into the last section, since it was (so far) meant to be used for visualization and understanding of the data.

Quite a lot of work to do …