An Algorithm for Incremental Mode Induction

Learning systems have been devised as a way of overcoming the knowledge acquisition bottleneck in the development of knowledge-based systems. They often cast learning to a search problem in a space of candidate solutions. Since such a space can grow exponentially, techniques for pruning it are needed in order to speed up the learning process. One of the biases used by Inductive Logic Programming (ILP) systems for this purpose is mode declaration. This paper presents an algorithm to incrementally learn this type of metaknowledge from the available observations, without requiring the final user's intervention.