文档介绍：©1999 Henrik Bengtsson
1
works - an introduction
Henrik Bengtsson
******@
Computer Science and Technology
Lund Institute of Technology
Master’s Thesis Project:
©1999 Henrik Bengtsson
2
Outline
• Introduction
• A simple work
• Graph Theory
• The Junction Tree Algorithm
• hbBN - a simple work tool
• The Twin-Model Example
• Summary
©1999 Henrik Bengtsson
3
Introduction
“The Year 2000 problem”
©1999 Henrik Bengtsson
4
History
‘60s The first expert systems. IF-THEN rules.
1968 Attempts to use probabilities in expert systems (Gorry & t).
1973 Gave up - to heavy calculations! (Gorry)
1976 MYCIN: Medical predicate logic expert system with certainty factors (Shortliffe).
1976 PROSPECTOR: Predicts the likely location of mineral deposits. Uses Bayes’ rule. (Duda et al.).
Summary of the time up until mid ‘80s:
• “Pure logic will solve the AI problems!”
• “Probability theory is intractable to use and plicated plex models.”
Example (PROSPECTOR):
P(d | a) = P(d | a b) ·P(b | a) + P(d | a b) ·P( b | a)
Certainty Factor (MYCIN):
A real value (-1,+1): -1: expression is known to be false. 0: no belief either way. +1: expression is known to be true.
Example:
rule 34: a b c q (+) rule 35: d q r s (-)
©1999 Henrik Bengtsson
5
But...
1986 works were revived and reintroduced to expert systems (Pearl).
1988 Breakthrough for efficient calculation algorithms (Lauritzen & Spiegelhalter) tractable calculations on BNs.
1995 In Windows95™ for printer-trouble shooting and Office assistance (“the paper clip”).
1999 BN is getting more and more used. Ex. Gene expression analysis, Business strategy etc.
2000 Widely used - a BN tool will be shipped with every Windows™ Commercial Server.
works is the future!
©1999 Henrik Bengtsson
6
a
b
c
d
A work has two parts:
1) qualitative part
• the structure
• directed acyclic graph (DAG)
• vertices represent variables
• edges represent relations
between variables
2) quantitati