CS62S: Expert Systems

1
CS62S Expert Systems

• Based on
• The Engineering of Knowledge-based Systems
  Theory and Practice,
• A. J. Gonzalez and D. D. Dankel

2
Rule-based systems (part 2)

• Backward reasoning
• Starting with the goal, rules are searched to see
  if the facts can lead to a value for the goal.

3
Backward reasoning steps

• Form a stack of all goals.
• Find all rules that support the first goal.
• Examine each rule
• If rule supports goal then remove goal from stack
  and go to step 2.
• If another rule is needed to satisfy the
  parameter of a rules premise then that is a
  subgoal and is placed on the stack
• Otherwise, ask the user for a value. If this
  value is not satisfactory, move on to the next
  rule.
• When the stack is empty, end execution.

4
Rule-based architecture

• Inference network
• Nodes facts, intermediate goals
• Rules connect the nodes
• All interconnections are known prior to execution

5

• Pattern-matching systems
• Relationships are formed during execution
• The premise of the rules are patterns, and may
  include
• Wild cards ? Single field, zero or more fields
• Logical operators NOT, OR
• Relational operators lt, gt
• Arithmetic operators

6

• Traditionally pattern-matching is used with
  forward chaining and inference net for backward
  chaining. However, either may be used with any
  reasoning method.

7
Rete Algorithm

• pattern-network
• Tree formed from the premise in the rules. Root
  is the first item in the premise.
• join network.
• Connects the tree nodes and compares similarly
  named variable for consistency in values

8
Knowledge-based systems lifecycle

• Waterfall model used in software engineering
• Problem analysis
• Requirements specification
• Design
• Implementation
• Testing
• Maintenance

9

• Problem analysis
• Requirements specification
• Preliminary design
• Initial prototype
• Detailed design
• Knowledge acquisition
• Validation and verification
• Design adjustment
• Maintenance

10
Feasibility Analysis

• Is there a need?
• Suitable for expert systems techniques?
• Is the knowledge heuristic?
• Does it mimic human problem solving methods?
• Does the knowledge change periodically?
• Is the knowledge well understood by experts?
• Are the input data always correct and complete?
• Can conventional programming solve it?
• Does it pass the telephone test?
• Is an expert system really justified?

11
Feasibility Analysis (cont)

• Resources
• Management support
• Available expert
• Competent
• Articulate
• Close proximity