Alessandro Antonucci reports on WPMSIIP 2013

Posted on January 7, 2014 by Alessandro Antonucci
[ go back to blog ]

The sixth edition of WPMSIIP, the Workshop on Principles and Methods of Statistical Inference with Interval Probability was held in Lugano (Switzerland) between the first and the second week of September 2013. The workshop was a follow-up to previous editions, held in Durham (2008, 2010), Munich (2009, 2012), and Lublijana (2011).

The 2013 edition was organized by the Imprecise Probability Group of IDSIA. About 25 participants from ten different countries attended the workshop. Since its first edition, WPMSIIP is intended as an open forum for researchers interested in interval (and imprecise, thus of great interest for the SIPTA community) probability. Almost all the participants took actively part to the workshop by presenting in their talks ongoing research topics and/or open challenges. Each talk was followed by an open discussion, with no strict time constraints. Yet, despite some very long and intense discussions, the original program was (of course in an imprecise way!) met.

During the first day, the discussion has been focused on classification and regression. Giorgio Corani chaired the classification part: it clearly emerged that the focus of the research on this topic is slightly moving from traditional credal classification to more general (and challenging) data-mining problems like preference learning and multilabel classification. The regression part was chaired by Andrea Wiencierz and the importance of new regression tools to cope with interval data was one of the main output of the discussion.

The second day, chaired by Alessio Benavoli and Marco Cattaneo, covered different topics related to learning. An increasing interest for filtering based on interval/imprecise methods was observed, together with the need of novel learning tools for non-parametric imprecise models.

Inference was the topic of the third day. Cassio Polpo de Campos chaired the discussion, which was mostly specialized to imprecise probabilistic graphical models. Two major topics were discussed: credal networks with epistemic irrelevance and the application to logic of imprecise models.

Decision making and evaluation problems were discussed on the fourth day, chaired by Denis Mauá. The utility-based approach seems a satisfactory answer to the evaluation of imprecise classifiers. The situation is definitely more open for decision making.

Finally, on the last day, open problems ranging from very theoretical to very research applied topics were discussed.

Despite such a packed program, it was possible to find time for a hiking excursion in the beautiful Swiss Alps. In summary, the WPMSIIPs meetings should be regarded as an important resource for the SIPTA community. The open format of the workshop allows for exhaustive (and sometimes exhausting!) discussions, something which could hardly be reproduced in other forums. Most of the slides of the talks are available in the workshop website. We look forward to seeing you at WPMSIIP 2014!