Relational learning with transfer of knowledge between domains

A commonly used relational learning system (FOIL) is extended through the use of cliches, which are known to address FOIL's greedy search deficiencies. The issue of finding good biases in the form of cliches is addressed by learning the cliches. This paper shows empirically that such biases can be learned in one domain and applied in another, and that significant improvement in accuracy can be achieved in this setting. The approach is applied to a real-life problem of learning finite Element method structures from examples.