Uniform Learning of Recursive Functions

Inductive inference is concerned with formal models of learning behaviour for the analysis of general learning phenomenons. In this context, the author studies uniform inductive inference - an approach to meta-learning in the sense that, given descriptions of solvable learning problems, a meta-learner has to develop successful learners for each of them. It turns out that uniform learning is not a trivial task: the success of meta-learners strongly depends on the choice of the particular descriptions of the learning problems. The present study investigates the capabilities of meta-learners, resulting from three different formal approaches to uniform learning. The strength and weakness of uniform learning is illustrated by numerous examples; in particular, the success of specific universal methods is analysed. Finally, a formal analysis of the complexity of different learning tasks concludes the study.