Representation and Reasoning With Uncertainty: Topic 4

Topic 4
Representation and Reasoning

with Uncertainty
Contents
4.0 Representing Uncertainty
4.1 Probabilistic methods (Bayesian)
4.2 Certainty Factors (CFs)
4.3 Dempster-Shafer theory
4.4 Fuzzy Logic
4.0 Representation and Reasoning with Uncertainty
Reasoning can be uncertain in several ways:

Uncertainty in data (facts)
Uncertainty in inference processes (rules)
Uncertainty in data (facts)

Data not available (e.g., doctor does not know if patient has
allergy to a drug.
Data unreliable: data is available, but not trustworthy, e.g.,
errors of measurement, two measurements gave
different results, etc.
The data is in an imprecise representation.
The data represents a guess by the expert,
e.g., Patient is in good health?
True (0.9)
The data may be just a default value, and there may be

exceptions to default values.
3
Uncertainty in knowledge (rules)

Rule may represent just a statistical association observed by
experts, e.g.,.
IF
primary infection is bacteriacea

AND site of infection is sterile
AND entry point of infection
is gastrointestinal tract
THEN organism is bacteroid (0.7)
Rule may not be appropriate in all cases

e.g., rules designed for the average person may not be suitable for
someone who is totally atypical
Issues in managing uncertainty

1. How to represent uncertain data and knowledge.
2. How to combine two or more items of uncertain data
(e.g., if both premises to a rule are uncertain, how do we
combine the uncertainy)
3. How to draw inferences (conclusions) using
uncertain data and knowledge (e.g., how do we
combine the uncertainty of premises and rules to
estimate the uncertainty of the conclusion)
Some ways for representing uncertainty:

1. As Probability (Bayesian Logic): what is the probability
of X being True
P(x)=0: X is false
P(x)=1: X is true
2. As a Certainty Factor: how certain are we that X is true

or not true.
CF(x)= 0: I have no knowledge if x is true or false
CF(x)= -1: I believe x is not true
CF(x)= 1: I believe x is true
3. As Fuzzy Logic: to what degree x is in various sets.

a given cup of coffee can be member of cold-drink and hot-drink
sets at the same time.
6
Topic 4
Representation and Reasoning
with Uncertainty

Definitions
Random Variable: an aspect of the problem for which a
value is not initially known.
Examples: patient temperature, patient has viral infection?
Domain: the range of possible values a variable can take.

Can be boolean (true/false), discreet (black/red/green) or
continuous values.
Atomic event (outcome): A complete specification of the
state of the world about which the agent is uncertain.
E.g., if the world consists of only two Boolean variables Cavity and
Toothache, then there are 4 distinct atomic events:
Cavity = false & Toothache = false
Cavity = false & Toothache = true
Cavity = true & Toothache = false
Cavity = true & Toothache = true
8
Example
Tossing a coin.
Random Variable X: the result of the toss
Domain: {head, tail}
Outcomes: {X=head, X=tail}
Two tosses of the same coin
Random Variable :
X (result of the first toss),
Y (result of the second toss)
Domain of both X and Y: {head, tail}
Outcomes: { {X=head, Y=head}, {X=head, Y=tail},
{X=tail, Y=head}, {X=tail, Y=tail} }
9
Example II
Patient being diagnosed with Flu base based on cough,
temperature, age, fever .
Random Variables and their domains:
Cough:
Temperature
{ yes, no}
10-45C
Age:
Fever:
Flu :
0-100
{yes, no}
{yes, no}
(boolean)
(discrete, if the thermometer has
resolution of 0.5 C)
(discrete)
(boolean)
(boolean)
10
More definitions
- Event: set of zero, one or various atomic events.
It represents something that can happen .
Example : In the problem involving two tosses of the coin,
the event "has at least one head" is the set :
{
{X=head, Y=head}, {X=head, Y=tail},

{X=tail, Y=head} }
The event the second toss is different to the first is:

{ {X=head, Y=tail}, {X=tail, Y=head} }.
11
More definitions
- Mutually exclusive events: events that do not have any
atomic event in common.
- This implies that they cannot happen simultaneously.
- Example: The event "the second toss is different to the first" and the
event "the second toss is the same as the first" are mutually
exclusive.
- The event there is at least one head and the event there is at least
one tail are not mutually exclusive.
- Exhaustive events: a set of events which together cover

all the atomic events of the problem.
12

Probabilities
The probability of event si is given as p(si)
The probability of each event si is
between 0 and 1
0<p(si)<1.0
If s is an event which must occur, p(s) = 1.

If s cannot occur, p(s) = 0
The sum of probabilities of all events is 1.0:
n
p(s ) = 1.0
i
i=1
The probability of ~s is 1- p(s):

p(~si) = 1 - p(si)
p(si) + p(~si) = 1
13

Combined Probability
If s1 and s2 are two outcomes (U = v = or; = & = = and) :
p(s1 s2) = p(s1) + p(s2) p(s1 s2)

e.g. p(dice<4) = 1/2, p(dice=even)=1/2,
p(dice<4 or dice=even) = 1/2 + 1/2 - 1/6 = 5/6
If s1 and s2 are mutually exclusive, this simplifies :
p(si v sj ) = p(si) + p(sj)
e.g.
p(dice=3) = 1/6, p(dice=4)=1/6,

p(dice = 3 or 4) = 2/6
14

Joint Probability
If xi and yj are independent events:
p(xi & yj ) = p(xi) p(yj)
e.g. p(dice1=3) = 1/6, p(dice2=4)=1/6,
p(dice1 = 3 & dice2=4) = 1/36
15

Conditional Probability
The probability of s1 given outcome s2:
p(s1| s2)
The conditional probability can be calculated
p(s1 | s2)
p(s1 s2)
p(s2)
E.g., p(dice=4 | dice=even)
= 1/6 / 1/2
= 1/3
given we roll an even dice, we have 1/3 chance of rolling a 4.

16

Conditional Probability with independent events
If x and y are totally independent events:
p( x & y) = p(x) . p(y)
Thus:
p( x | y) = p(x & y) / p(y)
= p(x) . p(y) / p(y)
= p(x)
17

Conditional Probability: alternative formula
Given:
p( y | x) = p(y & x) / p(x)
Then:
p(y & x) = p(y | x) * p(x)
Substituing this into our prior formula:

p( x | y)
= p(x & y) / p(y)
Gives:
p( x | y)
= p(y | x) * p(x) / p(y)
THIS IS CALLED BAYES THEOREM!!!

18

Bayes Rule: example
Given that p(dice=2|dice=even) = 1/3
Calculate p(dice=even|dice=2)
p(dice=even|dice=2) = p(dice=2|dice=even) * p(dice=even)

p(dice=2)
= 1/3 * 1/2
1/6
= 1/6
1/6
=1
19
10

Representation and Reasoning With Uncertainty: Topic 4

Uploaded by

Copyright:

Available Formats

Representation and Reasoning With Uncertainty: Topic 4

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Representation and Reasoning With Uncertainty: Topic 4

Uploaded by

Copyright:

Available Formats

Topic 4

Representation and Reasoning

4.0 Representation and Reasoning with Uncertainty

Reasoning can be uncertain in several ways:

4.0 Representation and Reasoning with Uncertainty

Uncertainty in data (facts)

The data may be just a default value, and there may be

4.0 Representation and Reasoning with Uncertainty

Uncertainty in knowledge (rules)

primary infection is bacteriacea

Rule may not be appropriate in all cases

4.0 Representation and Reasoning with Uncertainty

Issues in managing uncertainty

4.0 Representation and Reasoning with Uncertainty

Some ways for representing uncertainty:

2. As a Certainty Factor: how certain are we that X is true

3. As Fuzzy Logic: to what degree x is in various sets.

4.1 Probabilistic methods (Bayesian)

4.1 Probabilistic methods (Bayesian)

Domain: the range of possible values a variable can take.

4.1 Probabilistic methods (Bayesian)

4.1 Probabilistic methods (Bayesian)

4.1 Probabilistic methods (Bayesian)

{X=head, Y=head}, {X=head, Y=tail},

The event the second toss is different to the first is:

4.1 Probabilistic methods (Bayesian)

- Exhaustive events: a set of events which together cover

4.1 Probabilistic methods (Bayesian)

If s is an event which must occur, p(s) = 1.

The probability of ~s is 1- p(s):

4.1 Probabilistic methods (Bayesian)

p(s1 s2) = p(s1) + p(s2) p(s1 s2)

p(dice=3) = 1/6, p(dice=4)=1/6,

4.1 Probabilistic methods (Bayesian)

4.1 Probabilistic methods (Bayesian)

The conditional probability can be calculated

E.g., p(dice=4 | dice=even)

given we roll an even dice, we have 1/3 chance of rolling a 4.

4.1 Probabilistic methods (Bayesian)

4.1 Probabilistic methods (Bayesian)

p( y | x) = p(y & x) / p(x)

p(y & x) = p(y | x) * p(x)

Substituing this into our prior formula:

= p(x & y) / p(y)

= p(y | x) * p(x) / p(y)

THIS IS CALLED BAYES THEOREM!!!

4.1 Probabilistic methods (Bayesian)

p(dice=even|dice=2) = p(dice=2|dice=even) * p(dice=even)

You might also like