Challenge problems


To compare the systems we use a number of "challenge problems". A challenge problem can be given by one specific model that is encoded in a particular representation language (the "reference model"). Usually, these are model examples that come with the system implementing the language. The challenge for other languages/systems is to find an equivalent representation. Alternatively, a challenge can be formulated as a modeling task in general terms. The challenge problems are selected to be rather simple examples for fundamental modeling tasks.

Models marked "Leuven exercise" are contributed by Bruynooghe et (many!) al. from Leuven University. See the ILP 2009 paper "An Exercise with Statistical Relational Learning Systems".

  • Bloodtype model: A simple genetic model for the inheritance of bloodtypes in a pedigree. Input domains consist of pedigrees with individuals and "father" and "mother" relation. The reference model comes from Balios.

  • Reference model Complete list of models

  • University model: A model for the adviser relation between students and professors. Input domainsa consist of a number of students and professors. This model is representative for undirected models given by feature-induced potentials, rather than conditional probabilities. The reference model comes from Alchemy.

  • Reference model Complete list of models

  • HMM model: A standard HMM model. Input domains just consist of the number of time slices over which the model is to be developed (this need no neccessarily be explicitly specified as an extensional model part). We use several variants of an HMM model that differ in the number of states of the hidden variable, and investigate how this affects the representability of this model in languages that only permit binary variables (Primula and Alchemy).

  • Task description Complete list of models

  • Noisy-or model: A very basic model that uses the noisy-or function for combining sereral causal inputs. Input structures are directed acyclic graphs on which a random unary attribute is propagated from the roots downward. Reference model comes from Primula.

  • Reference model Complete list of models

  • ER-model (Leuven Exercise): An entity-relationship model in a real-estate agent/customer domain.

  • Modeling Task Lerning Task Data Complete list of models

  • Markov Models (Leuven Exercise): Different versions of a (hidden) Markov model.

  • Task description Data Complete list of models