Information Technology Reference
In-Depth Information
*INROOM (ROBOT, p2)
*CONNECTS (p3, p2, p5)
GOTHRU (p3, p2, p5)
*INROOM (p6, p5)
*CONNECTS (p8, p9, p5)
*INROOM (ROBOT , p5)
PUSHTHRU (p6, p8, p5, p9)
*CONNECTS (x, y, z)
*CONNECTS (x, z, y)
INROOM (ROBOT, p9)
INROOM (p6, p9)
Figure 9.6. Generalize triangle table
The learning method DeJong et al. used is to learn schemata. The main idea
of schemata is to arrange relevant knowledge into a group in order to attain a
certain objective while constructing systematical knowledge. Schemata are a
partial ordering set of operators, and the simple schemata relates to each other
together. EGGS applies dynamic operationality criterion, when the system learns
a general schema of a concept, this concept is operable. The problem of EGGS
can be depicted as(DeJong and Mooney, 1986):
Given
· Domain TheoryConsists of three parts: a) a specification of types of objects
in the world and their properties; b) a set of inference rules for inferring
properties and relations from other properties and relations; c) a library of
problem solving operators and already known general schemata. These
schemata can be previously learned or hand-coded.
· Goal: General specification of goal status.
· Initial world status: A specification of the objects in the world and their
properties.
· Observed operator/state sequence (optional): An observed sequence of
low-level operators, performed by an expert, which achieves an instance of
the goal. In some situations, some operators may be missing, in which case
they must be inferred from achieved state given in the input.
Determine
· A new schemata that achieves the goal status in a general way.
EGGS has always maintained two independent substitution tables SPECIFIC
and GENERAL. Table SPECIFIC is used for substitution of explanation structure
