EN FR
EN FR


Section: New Results

Machine learning for XML document transformations

Participants : Jérôme Champavère, Jean Decoster, Jean-Baptiste Faddoul, Antonino Freno, Gemma Garriga, Rémi Gilleron, Mikaela Keller, Grégoire Laurence, Aurélien Lemay, Joachim Niehren, Sławek Staworko, Marc Tommasi, Fabien Torre.

Induction of tree automata. Champavère, Gilleron, Lemay and Niehren proposed to use schemas for improving induction algorithms for monadic queries represented by tree automata [26] . The idea is to use pruning strategies to eliminate useless parts of trees when learning from partially annotated trees such that only the structure of relevant fragments is learned. This allows to avoid generalization errors and to learn from fewer annotations. They define schema-guided pruning strategies. They define stable queries w.r.t. a pruning strategy and show that stable queries are learnable.

Further Results. In [21] , Staworko proposed learning twig and path queries. Prioritized repairing and consistent query answering in relational databases was tackled in [13] and Bounded repairability for regular tree languages in [20] .

Torre and Terlutte explored the combination of automata and words balls for sequences classification in [14] .

Tommasi participated in the writing of a chapter of a book on conditional Markov fields for information extraction [23] .

Garriga and collaborators from the Fraunhofer Institute in Bonn, studied fixed parameter tractable algorithms for the discovery of maximal order preserving submatrices in bioinformatic applications in [17] .

We begun a new activity on learning for social network and information network supported by the arrivals of Gemma Garriga , Mikaela Keller and Antonino Freno . Freno, Garriga and Keller [22] proposed a model for predicting new links in a network which exploit both the current structure of the network and the content of its node.