PRODIGY: EBL

Description

• PRODIGY's EBL module performs the following steps in an effort to insure that performance in problem solving will increase as a result of explanation based learning.
  • Select a target concept - EBL can choose to learn any of the following concepts.
    • Why a path succeeded.
    • Why a path failed.
    • Certain other significant properties (such as why one path is shorter than another).
  • Compression - EBL attempts to reduce the explanation to its simplest form in order to reduce the cost of matching the resulting rule(s).
  • Utility - An explicit utility metric is used to evaluate the predicted usefullness of the resulting rule. Only rules with a high rate of utility are kept by the system. The following items are taken into account by the Utility Function:
    • The cost of matching a rule
    • The average savings when the rule is applied
    • The fraction of time that the rule is applicable when it is tested