Second Generation Expert Systems Ahme Rafea CS Dept., AUC Second - - PowerPoint PPT Presentation

Second Generation ES 1

Second Generation Expert Systems

Ahme Rafea CS Dept., AUC

Second Generation ES 2

First Generation ES Problems

• Knowledge Acquisition
• Explanation
• Brittleness
• KB Maintenance

Second Generation ES 3

Multiple Domain Knowledge Models

• Multiple Domain Models

– Different Representation – Deep Knowledge – Qualitative Reasoning – Different Level (e.g. diseases, anatomy,…) – Different Points of View(e.g. Structural and Functional)

Second Generation ES 4

Multiple Methods

• Specialized to a particular Representation
• Specialized for a particular subtask
• Combine multiple models and methods
• There are many criteria for choosing a

problem-solving architecture

• Domain Analysis is needed to choose the

problem-solving architecture

Second Generation ES 5

Knowledge Level

• Distinction between Knowledge Level and Implementation

Level

• Describing the problem-solving process in more abstract

terms such as classification, data abstraction, abduction, etc.

• Emphasizing how the problem will be solved rather that

how the system will be implemented

• Provides guidance in Knowledge Acquisition
• Reuse of Generic Constructs
• More abstract Explanation

Second Generation ES 6

KL Different Approaches

• Role Limiting Method, putting emphasis on

methods

• Generic Task is a method oriented

approach(Limited set of Primitive Constructs)

• KADS places emphasis on Task (general modeling

Language)

• Model Based Reasoning puts emphasis on Domain

Model

Second Generation ES 7

Second Generation Expert Systems

• Knowledge Acquisition
• Explanation
• Robustness and Efficiency
• Reusability

Second Generation ES 8

Knowledge Acquisition

• Modeling
• Instantiating the Model
• Validation
• Open Issues and On-going Work

– Development of the model and acquisition of domain knowledge may be interleaved – Selecting the appropriate model may not be easy

Second Generation ES 9

Explanation

• Explaining at the right level of abstraction

– The Task structure and the domain model provide the basis for more abstract explanation

• Re-constructive Explanation

– One KBS (performance system) is used to solve the problem, then conclusion of this system is passed to a second system (the explanation system) that justifies(and thus explains ) the conclusion

• Open Issues

– Producing explanation in natural language – Adapting the explanation to user needs – Managing the dialogue

Second Generation ES 10

Robustness and Efficiency

• Multiple Knowledge Sources

– Representing sufficient knowledge, improve the robustness of the system – Efficiency increases because only necessary knowledge is processed

• Multiple Representation

– Specialized representation, tuned to particular types of knowledge can make certain inferences very tractable

• Open Issues

– Difficulties in integrating diverse knowledge sources and representations

Second Generation ES 11

Reusability

• Reusing Knowledge-Level models
• Reusing Knowledge bases
• Reusing Symbol-level Components