Material used Frans Voorbraak: Certainty Factors (in the ... · Certainty Factors A Certainty...
Transcript of Material used Frans Voorbraak: Certainty Factors (in the ... · Certainty Factors A Certainty...
![Page 1: Material used Frans Voorbraak: Certainty Factors (in the ... · Certainty Factors A Certainty Factor is a numerical value that expresses the extent to which, based on a given set](https://reader030.fdocuments.net/reader030/viewer/2022021610/5c89147309d3f2bb368ca7a7/html5/thumbnails/1.jpg)
Certainty Factor Models (CFM)
Material used Frans Voorbraak: Certainty Factors (in the reader,
lecture)
![Page 2: Material used Frans Voorbraak: Certainty Factors (in the ... · Certainty Factors A Certainty Factor is a numerical value that expresses the extent to which, based on a given set](https://reader030.fdocuments.net/reader030/viewer/2022021610/5c89147309d3f2bb368ca7a7/html5/thumbnails/2.jpg)
2
Methods for Dealing with Uncertainty
• 1. Numeric Methods – Bayesian Model – Certainty Factor Model – Dempster - Shafer Theory – Possibility Theory
• 2. Fuzzy Logic
• 3. Non - Monotonic Logic
![Page 3: Material used Frans Voorbraak: Certainty Factors (in the ... · Certainty Factors A Certainty Factor is a numerical value that expresses the extent to which, based on a given set](https://reader030.fdocuments.net/reader030/viewer/2022021610/5c89147309d3f2bb368ca7a7/html5/thumbnails/3.jpg)
3
Certainty Factors
A Certainty Factor is a numerical value that
expresses the extent to which, based on a given set of evidence, we should accept a given conclusion. A Certainty Factor or CF with a value of 1 indicates total belief, whereas a CF with a value of -1 indicates total disbelief.
In a system that uses CFs, the rules must be so structured that any given rule either adds to belief in a given conclusion or adds to disbelief.
![Page 4: Material used Frans Voorbraak: Certainty Factors (in the ... · Certainty Factors A Certainty Factor is a numerical value that expresses the extent to which, based on a given set](https://reader030.fdocuments.net/reader030/viewer/2022021610/5c89147309d3f2bb368ca7a7/html5/thumbnails/4.jpg)
4
Elementary illustration
&0.6
&0.5
0.90.95
0.4+0.54-0.4x0.54 =0.72
0.9x0.6=0.50.8x0.5=0.4
Influenza
Cough Fever FeverPain0.8 0.9
![Page 5: Material used Frans Voorbraak: Certainty Factors (in the ... · Certainty Factors A Certainty Factor is a numerical value that expresses the extent to which, based on a given set](https://reader030.fdocuments.net/reader030/viewer/2022021610/5c89147309d3f2bb368ca7a7/html5/thumbnails/5.jpg)
5
General characteristics of the CFM
![Page 6: Material used Frans Voorbraak: Certainty Factors (in the ... · Certainty Factors A Certainty Factor is a numerical value that expresses the extent to which, based on a given set](https://reader030.fdocuments.net/reader030/viewer/2022021610/5c89147309d3f2bb368ca7a7/html5/thumbnails/6.jpg)
6
Certainty Theory
• Representing uncertain evidence (facts) • Representing uncertain rules • Combining evidence from multiple sources
eg. IF A and B THEN X [CF= 0.8] IF C THEN X [CF=0.7] What is the certainty of X? (given the CFs of A,B,C)
• -1<= CF <= 1 • Users provide CFs for specific facts; Experts
provide CFs for rules • Explanation of CFs in terms of Probabilities?
![Page 7: Material used Frans Voorbraak: Certainty Factors (in the ... · Certainty Factors A Certainty Factor is a numerical value that expresses the extent to which, based on a given set](https://reader030.fdocuments.net/reader030/viewer/2022021610/5c89147309d3f2bb368ca7a7/html5/thumbnails/7.jpg)
7
Some notations
CF(A) CF of a uncertain evidence
CF(IF A THEN B) CF of a rule
also CF(A B) CF(X|R) CF of X given the rules in
the set R (+ required facts) X is a non-initial fact
![Page 8: Material used Frans Voorbraak: Certainty Factors (in the ... · Certainty Factors A Certainty Factor is a numerical value that expresses the extent to which, based on a given set](https://reader030.fdocuments.net/reader030/viewer/2022021610/5c89147309d3f2bb368ca7a7/html5/thumbnails/8.jpg)
8
Inference network
![Page 9: Material used Frans Voorbraak: Certainty Factors (in the ... · Certainty Factors A Certainty Factor is a numerical value that expresses the extent to which, based on a given set](https://reader030.fdocuments.net/reader030/viewer/2022021610/5c89147309d3f2bb368ca7a7/html5/thumbnails/9.jpg)
9
Calculations
![Page 10: Material used Frans Voorbraak: Certainty Factors (in the ... · Certainty Factors A Certainty Factor is a numerical value that expresses the extent to which, based on a given set](https://reader030.fdocuments.net/reader030/viewer/2022021610/5c89147309d3f2bb368ca7a7/html5/thumbnails/10.jpg)
10
Propagation rules
![Page 11: Material used Frans Voorbraak: Certainty Factors (in the ... · Certainty Factors A Certainty Factor is a numerical value that expresses the extent to which, based on a given set](https://reader030.fdocuments.net/reader030/viewer/2022021610/5c89147309d3f2bb368ca7a7/html5/thumbnails/11.jpg)
11
Motivation of the combination rules
Antecedent pooling: from fuzzy logic Serial combination: the factor max(0,CF(A)) avoids
that evidence against A can have an effect on the certainty of B through A B
The parallel combination function f is an improved version of the following combination function which was originally used:
(note that it is undefined for x = 1, y= -1, and for x = -1, y=1)
![Page 12: Material used Frans Voorbraak: Certainty Factors (in the ... · Certainty Factors A Certainty Factor is a numerical value that expresses the extent to which, based on a given set](https://reader030.fdocuments.net/reader030/viewer/2022021610/5c89147309d3f2bb368ca7a7/html5/thumbnails/12.jpg)
12
Illustrating example
R1: IF weatherman says it will rain THEN it will rain CF 0.8 R2: IF farmer says it will rain THEN it will rain CF 0.8 Case (a): Weatherman and farmer are certain in rain CF(E1) = CF(E2) = 1.0 CF(H, E1) = CF(E1) * CF(Rule1) = 1.0*0.8 = 0.8 CF(H, E2) = CF(E2) * CF(Rule2) = 1.0*0.8 = 0.8 CFcombine(CF1, CF2) = CF1+ CF2(1- CF1) = 0.8+0.8(1-0.8) =
0.96 CF of a hypothesis which is supported by more than one rule,
can be incrementally increased by supporting evidence from both rules.
![Page 13: Material used Frans Voorbraak: Certainty Factors (in the ... · Certainty Factors A Certainty Factor is a numerical value that expresses the extent to which, based on a given set](https://reader030.fdocuments.net/reader030/viewer/2022021610/5c89147309d3f2bb368ca7a7/html5/thumbnails/13.jpg)
13
Illustrating example, continued
Case (b):Weatherman certain in rain, farmer almost certain in no rain
CF(E1) = 1, CF(E2) = -0.99 CF1 = 0.8, CF2 = -0.792 CFcombine(CF1, CF2) = (0.8 + (-0.792)) / (1 - min(0.8,
0.792) = 0.04
Case (c): Incremental decrease in certainty from more than one source of disconfirming evidence
CFcombine(CF1, CF2, CF3,….) = 0.999 = CFold Single piece of disconfirming evidence CFnew = -0.8 Cfcombine = (0.999 - 0.8) / (1 - 0.8) = 0.995 Single piece of disconfirming evidence does not have a
major impact on many pieces of confirming evidence.
![Page 14: Material used Frans Voorbraak: Certainty Factors (in the ... · Certainty Factors A Certainty Factor is a numerical value that expresses the extent to which, based on a given set](https://reader030.fdocuments.net/reader030/viewer/2022021610/5c89147309d3f2bb368ca7a7/html5/thumbnails/14.jpg)
14
Problems with CFs
![Page 15: Material used Frans Voorbraak: Certainty Factors (in the ... · Certainty Factors A Certainty Factor is a numerical value that expresses the extent to which, based on a given set](https://reader030.fdocuments.net/reader030/viewer/2022021610/5c89147309d3f2bb368ca7a7/html5/thumbnails/15.jpg)
15
Problems with CFs, continued
![Page 16: Material used Frans Voorbraak: Certainty Factors (in the ... · Certainty Factors A Certainty Factor is a numerical value that expresses the extent to which, based on a given set](https://reader030.fdocuments.net/reader030/viewer/2022021610/5c89147309d3f2bb368ca7a7/html5/thumbnails/16.jpg)
16
Foundation
Suppose a domain expert believes with 70 % certainty that gram positive cocci growing in chains are streptococci.
What is an appropriate measure of belief to represent this knowledge? MB(E H) = P(H|E) ?? Problem: MB(E H) = 0.7, then MB(E ∼H) = 0.3
![Page 17: Material used Frans Voorbraak: Certainty Factors (in the ... · Certainty Factors A Certainty Factor is a numerical value that expresses the extent to which, based on a given set](https://reader030.fdocuments.net/reader030/viewer/2022021610/5c89147309d3f2bb368ca7a7/html5/thumbnails/17.jpg)
17
Foundation, continued
1 if P(H) = 1
MB(E H) = )(1)()|(
HPHPEHP
−−
if P(H|E)
0 otherwise CF(E H) = MB(E H) - MB(E ∼H)
The Measure of Belief (MB) is a number that reflects the measure of increased belief in a hypothesis H based on evidence E. (Shortliffe & Buchanan)
![Page 18: Material used Frans Voorbraak: Certainty Factors (in the ... · Certainty Factors A Certainty Factor is a numerical value that expresses the extent to which, based on a given set](https://reader030.fdocuments.net/reader030/viewer/2022021610/5c89147309d3f2bb368ca7a7/html5/thumbnails/18.jpg)
18
Foundation, continued
Obvious consequence:
So far, CFP is only defined for rules. In order to apply it to facts we can define CFP(A) = CFP(e A) where e is the external evidence available to the user.
![Page 19: Material used Frans Voorbraak: Certainty Factors (in the ... · Certainty Factors A Certainty Factor is a numerical value that expresses the extent to which, based on a given set](https://reader030.fdocuments.net/reader030/viewer/2022021610/5c89147309d3f2bb368ca7a7/html5/thumbnails/19.jpg)
19
Foundation of propagation?
It can be shown that, given the above probabilistic interpretation of CF’s , the propagation rules of the CT model are only valid under trivializing assumptions. Perhaps they are reasonable approximations in some cases. The following example shows that this is a rather weak line of defence.
![Page 20: Material used Frans Voorbraak: Certainty Factors (in the ... · Certainty Factors A Certainty Factor is a numerical value that expresses the extent to which, based on a given set](https://reader030.fdocuments.net/reader030/viewer/2022021610/5c89147309d3f2bb368ca7a7/html5/thumbnails/20.jpg)
20
Counterexample
Let Ω = 1,2,3,4,5,6. Then define for every X ⊆ Ω, P(X) = |X|/6.
Assume the user’s evidence E is represented by 1,2. Further, assume A is represented by 1 and B is represented by 2.
Then CFP(A) = CFP(E A) = (1/2 – 1/6)/(1-1/6) = 0.4. Similarly, CFP(B) = 0.4.
If CFP(A) = CFP(B) = 0.4, then according to the propagation rules CF(A∧B) = CF(A∨B) = 0.4.
However, CFP(A∧B) = -1 and CFP(A∨B) = 1.
![Page 21: Material used Frans Voorbraak: Certainty Factors (in the ... · Certainty Factors A Certainty Factor is a numerical value that expresses the extent to which, based on a given set](https://reader030.fdocuments.net/reader030/viewer/2022021610/5c89147309d3f2bb368ca7a7/html5/thumbnails/21.jpg)
21
Evaluation