The 11th International Symposium on Intelligent Data Analysis (IDA 2012) is on for the final day. The prestigious Finlandia hall has been offering a great venue, and the organizers even ordered first snow to fall during the conference.
The keynotes have been on various subjects with some overlap in themes. Both Paola Sebastiani and Arno Siebes shared the view that offering a single value as the end result could be improved by offering a series of values which identify the system. With Arno Siebes this was a series of models with increasing complexity in his framework to explain different data sets with querying instead of tabulated data. The series of models could then be used to select the wanted accuracy and complexity without overfitting. Paola Sebastiani explain how they create profiles in their research on exceptional longevity as a series of prediction accuracies with an embedded sets of identified SNPs. Gavin C. Cawley is currently talking about over-fitting in model selection, whereas it is typically considered as a problem in parameter estimation in the first-level inference.