The workshop runs from Thursday the 5th of September until Tuesday the 10th of September 2013. Each of the five workdays of the workshop starts at 9am and will be focused on a specific topic. Here below some information about the topics of each day and a preliminary program.


This day the discussion will be focused on CLASSIFICATION and REGRESSION with imprecise models. The problem of learning multivariate models from incomplete and/or imprecise data will receive a special attention . The discussion will be chaired by Giorgio Corani for the classification part and Andrea Wiencierz for the regression part.
[talk #1] Giorgio CoraniClassification and imprecise probabilities: overview and open problems
[talk #2] Gen YangBinary decomposition for classification
[talk #3] Paul FinkClassification with decision trees
[talk #4] Andrea MignattiCredal model averaging of logistic regression
[talk #5] Andrea WiencierzRegression and imprecise probabilities: overview and open problems
[talk #6] Georg SchollmeyerDescriptive and inferential characteristics for set-valued regression-estimators
[talk #7] Andrea WiencierzA LIR method for SVR

FRIDAY (6 September) - LEARNING

On the second day the discussion will be about general LEARNING (intended as statistical inference) issues within the framework of imprecise probability. The problems of modeling prior ignorance and learning from incomplete data will receive a special attention. The discussion will be chaired by Alessio Benavoli and Marco Cattaneo.
[talk #1] Alessio BenavoliLearning imprecise probabilities: overview and open problems
[talk #2] Alessio BenavoliNonparametric estimation. An imprecise look to old ideas.
[talk #3] Gert de CoomanTrying to characterise the IDM, representation invariance and specificity
[talk #4] Marcio Alvez DinizRelationship between the Fundamental Theorem of Probability and de Finetti's Representation Theorem
[talk #5] Fabrizio RuggeriSome issues in Bayesian robustness
[talk #6] Marco CattaneoContinuous updating rules for imprecise probabilities
[talk #7] Matthias TroffaesKalman filtering, non-Bayes updating, and variational assimilation in high-resolution wind prediction: a practical perspective
[talk #8] Alessio BenavoliSome issues in HMM estimation


On the third day INFERENCE (intended as probabilistic inference) in imprecise-probabilistic models will be discussed. A special focus will be given to graphical models. The discussion will be chaired by Cassio de Campos.
[talk #1] Cassio de CamposInference with imprecise probabilities: overview and open problems
[talk #2] Jasper de BockCredal networks under epistemic irrelevance
[talk #3] Denis Maua'Inference algorithms for credal networks
[talk #4] David HuberCREDO: a software for credal networks
[talk #5] Alessandro AntonucciProbabilistic logic and credal networks
[talk #6] Arthur Van CampModelling practical certainty and its link with classical propositional logic
[talk #7] Matthias TroffaesModel checking for infinite horizon Markov chains


We will walk from the Bellavista station (see here) of the cogwheel train to the top station of the train (path number 2 above). Uphill it takes around 1.30 hours (400m ascent to the restaurant + 100m to get to the very top of the mountain). Downhill, it takes 1 hour. The top is at 1700m and it's often windy: warm clothes are recommended. People can also take the train to the top.
We suggest to bring a warm sweater, windbreaker or coat, sun cream, sun hat, sunglasses, and water. This is enough if you take the train. If you walk we also recommend a proper footwear. The path is easy: hiking boots are suggested but not strictly necessary. Hiking shoes are OK with dry weather. Comfortable (running/tennis) shoes may be also OK with dry weather. Food and drinks are available at the restaurant on the top. Still, one may bring its own food/drinks and consume it outside of the restaurant building.


This day the discussion will be about the performance EVALUATION of imprecise approaches and DECISION making based on these models. Starting from the case of credal classifiers, the general problem of comparing imprecise methods among them and with classical ones will be discussed. The day will be chaired by Denis Maua'. Note that on this day the workshop will be held at IDSIA.
[talk #1] Marco CattaneoThe meaning of robustness
[talk #2] Giorgio CoraniUtility-based evaluation of credal classifiers
[talk #3] Cassio de CamposImprecise probability and maximum entropy
[talk #4] Sebastien DesterckeInference and decision in multilabel and label ranking problems
[talk #5] Jasper de BockDecision making in credal networks
[talk #6] Alessandro AntonucciTaking E-admissible decisions

TUESDAY (10 September) - OPEN TOPICS

This day the discussion will be on OPEN TOPICS and chaired by Alessandro Antonucci.
[talk #1] Enrique MirandaConglomerability and coherent lower previsions
[talk #2] Alice CicirielloImprecise probability models in structural dynamics
[talk #3] Sebastien DesterckeSome thought about learning and IP models