Information Technology Reference
In-Depth Information
intuitionistic advice is given according to outer instruction from experts; they
should be similar to the initial problems. Sub-goal is created to solve this
problem and the sequences of operators are obtained. The learning mechanism
learns the memory chunk based on each sub-goal solving therefore the system
can solve the initial problem from memory chunk without asking for guidance.
The memory chunk, which uses working-memory-elements to collect
conditions and constructs memory chunk in SOAR system, is the key for learning.
When a sub-goal is created for solving a simple problem or assessing the advice
from experts, the current statuses are stored into w-m-e. System gets initial
statuses of sub-goal from w-m-e and deletes solution operators as the conclusion
action after the sub-goal is solved. This generative production rule is memory
chunk. If the sub-goal is similar to the sub-goal of the initial problem, memory
chunk can be applied to initial problem and the learning strategy can apply what
has already learned from one problem to another.
The formation of memory chunk depends on the explanation of sub-goal. The
imparting learning is applied when converting the instructions of experts or
simple problems into machine executable format. Lastly, experiences obtained
from solving simple and intuitionistic problems can be applied to initial problems,
which involve analogy learning. Therefore, the way of learning in SOAR system
is a comprehensive combination of several learning methods.
9.8 Operationalization
Operationalization is a process of translating a non-operational expression into an
operational one. The initial expression might represent a piece of advice or
concept which arenonoperational with respect to an agent because they are not
expressed in terms of data and actions available to the agent. An operationalizing
program faces the task of reformulating the original expression in terms of data
and actions that are available to the agent.Operationality criterion is one of the
input requirements of EBG. As for operationality criterion, Mitchell, Keller and
Kedar-Cabelli pointed out“concept description should be represented as
predicates of training examples and predicates selected from domain theory and
easier to assess” (Mtchell et al., 1986). Obviously, this is intuitionistic and simple,
only one predicate operational (pid) should be added into domain theory and
explain what predicates are operable. However, this kind of processing is static
which can not satisfy the demand of practical learning system. Therefore more
importance has been attached to how to define the operationality criterion as
dynamic, such as introducing theorem prover mechanism to define the
operationality criterion dynamically, EBG can even be adopted to change
operationality criterion. In this way, inference mechanism should be introduced
