Learning structured models for string and tree structured data
In this habilitation thesis, we review some approaches for learning structured models for string or structured data. We consider three particular subjects. In a first part, we address the problem of learning stochastic models represented by weighted automata. In a second part, we present two methods based on an oracle. One for learning rational functions over trees, the other for learning a powerful subclass of context-free grammars. Finally, we present some algorithms for learning similarities based on edit distance for string and trees.