Jump to the PRODIGY Overview page.
Jump to the Comparative Properties page.
Jump to the Home page.

Properties

Knowledge Representation

Knowledge in PRODIGY is represented in PDL, a form of the first-order logic. This representation is shared by the planner/problem solver and all of the learning modules. PRODIGY's prodecural knowledge is in the form of operators and control rules.

PRODIGY employs a glass box approach to knowledge. All knowledge is global, and can be examined by all the components of the architecture. Thus, knowledge acquired by one module can be used by the others.

Learning

Learning in PRODIGY is deliberative. PRODIGY attempts to learn when it the planner lacks control rules to help it solve a given problem. Also, based on the utilities of control rules, PRODIGY decides which rules to forget. The approach adopted in the architecture are casual commitment and deliberative learning strategies. The EBL module determines the utility of the control rules it generates.

Organization

The archtitecture is highly modular. Learning is handled by six separate modules. The modules do not interact directly. Knowledge is shared in a global database, and a module only influences another module's activity indirectly through changes to the database. Learning in PRODIGY is non-monotonic. Control rules may be deleted from the knowledge base as the estimate of their utility changes through experience.

Performance

PRODIGY's behavior entails a great deal of deliberation. Problem solving is not done in real-time in PRODIGY.