Learning methods such as neural network and decision tree need a certain number of training samples to achieve a certain level of generalization accuracy. Analytic learning uses prior knowledge and deductive reasoning to expand the information provided by training examples, so it is not restricted by the same boundaries. This paper introduces an analytic learning method called explanation based learning (EBL). In interpretive learning, prior knowledge is used to analyze how the observed learning examples satisfy the goal concept. This explanation is then used to distinguish between the relevant and unrelated features in the training samples. In this way, the examples can be generalized based on logical reasoning rather than statistical reasoning. This interpretation can make the learners have higher accuracy than relying on data alone. Starting from Prolog-EBG, this paper first introduces the general characteristics of this algorithm and the relationship between other inductive learning algorithms. Finally, the application of interpretive learning to improve the performance of large state space search is described.
Gadbois D , Miranker D P . Discovering Procedural Executions of Rule-Based Programs[C]// Proceedings of the 12th National Conference on Artificial Intelligence, Seattle, WA, USA, July 31 - August 4, 1994, Volume 1. American Association for Artificial Intelligence, 1994.