A Preliminary Study Of Inductive Learning Under Attributional Calculus

Artificial intelligence studies the mechanism of intelligent behavior.It studies intelligence by constructing and evaluating artifacts with intelligent behavior.According to the environment data,the machine learning system obtains the method to deal with a certain problem through certain learning methods.In the field of machine learning,inductive learning refers to the process in which an agent obtains a general concept,rule or theory through a certain search method according to the symbolic data described by a given description language.Attributional Calculus is a descriptive language that can represent inductive learning process.It is introduced by computer scientist Michalski.It is formalized on the basis of "attribute and attribute value have relationship",and gives a variety of evaluation scheme more than binary logic.The algorithm is to solve the problem,so the algorithm is the process description of solving the problem.When the inductive learning process focuses on a certain kind of inductive learning problem,it can be described by inductive learning algorithm.The inductive learning algorithm defines two opposite operators,generalization and specialization,on the basis of tautological implication,and then defines generalization rules.Based on the combination of Attributional Calculus and generalization rules,this paper attempts to apply AC to a simple candidate elimination algorithm,so as to explore the possibility of combining inductive learning with artificial neural network under the guidance of solving the difficult interpretability of artificial neural network.Reflections on inductive learning of Attributional Calculus show that for generalizers and other symbols,it has induction,attribute relationship,operator and function interpretation.There are many ways to interpret a symbol.Logic language based on computer practice has the advantages of strong expressiveness and the disadvantages of fuzziness.This is especially evident in the comment of inductive learning of Attributional Calculus representation.In the transformation of attribution calculus to other forms,attribution theory transforms the expression of individual sets and attributes,which makes it possible for Attributional Calculus sentences based on the relationship between attributes and attribute values to be transformed into attribution form,so that Attributional Calculus can also be transformed into set theory.The first-order predicate calculus is directly related to inductive learning.The Attributional Calculus sentence can be transformed into a normal form.By adding positive and negative example predicates,the problem can be expressed more simply.Then,how to reshape inductive learning under the guidance of interpretable artificial intelligence is a problem.In comparison with other research directions of inductive learning and machine learning,the ANN model deals with learning data in numerical domain,which is in contrast with the Attributional Calculus in symbol domain.They have their own advantages and disadvantages.The disadvantage of the representation of Attributional Calculus lies mainly in its discrete structure and complex computational space.It is a future trend that it can learn from the artificial neural network.One direction is to establish the hypothesis of human intelligence model based on synthetic of logic method and artificial neural network learning.Another direction is based on inductive logic programming.Inductive logic programming is based on first-order predicate logic.Its problem representation and inductive method are similar to that of Attributional Calculus.The higher order,probability and differentiability of inductive logic programming are contrasted with other explanatory modes of Attributional Calculus.