Information Technology Reference
In-Depth Information
Chapter 7
Inductive Learning
7.1 Introduction
Inductive learning is one of the most extensive approaches in symbol learning. Its
task is to induct a general conceptual description from a series of known positive
and negative examples given about a concept. Through inductive learning, new
concepts can be obtained, new rules created and new theories found.
Generalization and specification are the general operations of inductive learning.
Generalization is used to expand assumed semantic information so that it can
include more positive examples and be applied in more situations. Specialization
is an opposite operation of generalization. It is used to constrain application range
of conceptual description.
Inductive learning program is the procedure to describe the content
mentioned above by programming language. The language to write the inductive
program is referred to as inductive programming language. The system which
can execute inductive program and accomplish the specialized task of inductive
learning is referred to as inductive learning system, which could be independent
or embedded into another greater knowledge processing system. The input of
general inductive program is the description of few observations in scientific
experiments, while the output is the overall feature description of an object
category or classification discriminative description of several object categories.
In contrast to deduction, the start premise of induction is concrete fact rather
than general truth while the reasoning objective is likelihood general assertion to
explain fact in form and predict new truth. Induction attempts to lead to complete
and correct description from given phenomenon or part of its concrete
observations. Induction has two aspects − generating likelihood assumption and
its effectiveness (construction of truth value status). Only the former has
preliminary significance of the research of inductive learning. The assumption
effectiveness is secondary, because the assumption generated by hypothesis is
260
