| Popis: |
Several inductive learning algorithms focussing on the generalization of theories in first order logic and on the invention of new predicates have evolved in the last few years. We start with a brief introduction to the methods published by Muggleton & Buntine [5], Banerji [1, 2], Ling [3], Rouveirol & Puget [8], and Muggleton [4]. We then provide a critical and comparative analyses of these methods, pointing out some of their advantages and limitations. We show that each of these methods is a special form of the general theory of inverse resolution. In addition, we suggest some possible extensions to these methods. |
