Information Technology Reference
In-Depth Information
Chapter 9
Explanation-Based Learning
Explanation based learning (abbreviated as EBL) is a kind of analytic learning.
Under the guidance of domain theory, this learning method constructs the
explanation structure of cause and effect of the solving process based on the
analysis of only one single example solving and acquires control information
which can be used for solving similar problems later.
9.1 Introduction
Explanation-based learning was originally proposed by DeJong from the
University of Illinois in 1983. On the basis of empirical inductive learning, EBL
takes advantage of domain theory to explain single problem solving. This is an
inference analysis about cause and effect of knowledge and general control
strategy can be generated.
In 1986 Mitchell, Keller and Kedar-Cabelli put forward a unified framework
of explanation based generalization (abbreviated as EBG) and defined the EBL as
the following two steps(Mitchell et al., 1986):
(1) Generate an explanation structure by means of analysis one problem solving.
(2) Generalize the explanation structure and obtain general control rules.
DeJong and Mooney proposed a more general terminology EBL and it
became an independent branch of machine learning. As a kind of deductive
learning essentially, EBL makes deductive inference, stores useful conclusions
and constructs control information for future similar problem solving after
knowledge refinement and compilation. Unlike empirically-based learning
methods that require large numbers of examples, EBL always investigates one
training instance (commonly positive instance) deeply to get an explanation
structure. The analysis consists of three steps: first interprets why the training
instance is an example one of the desired concepts (goal concepts) and then
328
Search WWH ::




Custom Search