Acting Under Uncertainty Basic Concepts

Plan
 Why bother?
 Two examples
 Basic probability theory
 Bayes’ rule
 Bayes’ nets
 The Bayesian brain
 Reading: Russell &Norvig pp.462-488
AI Principles, Lecture on Reasoning Under Uncertainty

Why do we need reasoning under uncertainty?
 There are many situations where uncertainty arises:
• When you travel you reason about the possibility of delays

• When an insurance company offers a policy it has calculated the risk that
you will claim
• When your brain estimates what an object is it filters random noise and fills
in missing details
• When you play a game you cannot be certain what the other player will do
• A medical expert system that diagnoses disease has to deal with the results
of tests that are sometimes incorrect
 Systems which can reason about the effects of uncertainty

should do better than those that don’t
 But how should uncertainty be represented?

Probability Theory: Variables and Events
 A random variable can be an observation, outcome or event the value of
which is uncertain.
 e.g a coin. Let’s use Throw as the random variable denoting the
outcome when we toss the coin.
 The set of possible outcomes for a random variable is called its domain.
 The domain of Throw is {head, tail}
 A Boolean random variable has two outcomes.
 Cavity has the domain {true, false}
 Toothache has the domain {true, false}

Probability Theory: Atomic events
 We can create new events out of combinations of the outcomes of
random variables
 An atomic event is a complete specification of the values of the random

variables of interest
 e.g. if our world consists of only two Boolean random variables, then the
world has a four possible atomic events
Toothache = true ^ Cavity = true

Toothache = true ^ Cavity = false
Toothache = false ^ Cavity = true
Toothache = false ^ Cavity = false
 The set of all possible atomic events has two properties:

• It is mutually exhaustive (nothing else can happen)
• It is mutually exclusive (only one of the four can happen at one time)

Probability theory: probabilities
 We can assign probabilities to the outcomes of a random variable.
P(Throw = heads) = 0.5
P(Mary_Calls = true) = 0.1
P(a) = 0.3
 Some simple rules governing probabilities
1. All probabilities are between 0 and 1 inclusive 0  P(a)  1

P (true)  1
2. If something is necessarily true it has probability 1
P ( false)  0
3. The probability of a disjunction being true is
P (a  b)  P (a )  P (b)  P (a  b)
From these three laws all of probability theory can be derived.

Probability theory: relation to set theory
 We can often intuitively understand the laws of probability
by thinking about sets
P (a  b)  P (a )  P (b)  P (a  b)

Probability Theory: Conditional Probability
 A conditional probability expresses the likelihood that one event a will
occur if b occurs. We denote this as follows
P ( a | b)
 e.g.
P (Toothache  true)  0.2
P (Toothache  true | Cavity  true)  0.6
 So conditional probabilities reflect the fact that some events make other
events more (or less) likely
 If one event doesn’t affect the likelihood of another event they are said
to be independent and therefore
P ( a | b)  P ( a )
 E.g. if you roll a 6 on a die, it doesn’t make it more or less likely that you
will roll a 6 on the next throw. The rolls are independent.
Combining Probabilities: the product rule
 How we can work out the likelihood of two events occuring
together given their base and conditional probabilities?
P (a  b)  P (a | b) P (b)  P (b | a ) P (a )
 So in our toy example 1:
P (toothache  cavity)  P (toothache | cavity) P (cavity)

 P (cavity | toothache) P (toothache)
 But this doesn’t help us answer our question:

“I have toothache. Do I have a cavity?”

Bayes’ rule
 We can rearrange the two parts of the product rule:
P (a  b)  P (a | b) P (b)  P (b | a ) P (a )
P (a | b) P (b)  P (b | a ) P (a )
P (b | a ) P (a )
P ( a | b) 
P (b)
 This is known as Bayes’ rule.
 It is the cornerstone of modern probabilistic AI.
 But why is it useful?

Bayes’ rule
 We can think about some events as being “hidden” causes: not
necessarily directly observed (e.g. a cavity).
 If we model how likely observable effects are given hidden causes (how
likely toothache is given a cavity)
 Then Bayes’ rule allows us to use that model to infer the likelihood of the
hidden cause (and thus answer our question)
P (effect | cause) P (cause)

P (cause | effect ) 
P (effect )
 In fact good models of P (effect | cause ) often available to us in real

are
domains (e.g. medical diagnosis)

Bayes’ rule can capture causal models
 Suppose a doctor knows that a meningitis causes a stiff neck in 50% of cases
P ( s | m)  0.5
 She also knows that the probability in the general population of someone having a
stiff neck at any time is 1/20
 0.05 in the population (1/50,000)

P (ofs )meningitis
 She also has to know the incidence
 Using Bayes’ rule she can calculate the probability the patient has meningitis:
P (m)  0.00002
P ( s | m) P (m) 0.5  0.00002

P(m | s)    0.0002  1 / 5000
P( s) 0.05
P (effect | cause) P (cause)
P (cause | effect ) 
P (effect )
The power of causal models
 Why wouldn’t the doctor be better off if she just knew the likelihood of
meningitis given a stiff neck? I.e. information in the diagnostic direction
from symptoms to causes?
 Because diagnostic knowledge is often more fragile than causal

knowledge
 Suppose there was a meningitis epidemic? The rate of meningitis goes

up 20 times within a group
P ( s | m) P (m) 0.5  0.0004

P(m | s)    0.004  1 / 250
P( s) 0.05
 The causal model is unaffected by the change in P(m), whereas the
diagnostic model P(m|s)=1/5000 is now badly wrong.

Bayes rule: the normalisation short cut
 If we know P(effect|cause) for every cause we can avoid having to know
P(effect)
P (e | c ) P (c ) P (e | c ) P (c )
P (c | e)  
P (e)  P (e | h ) P ( h )
hCauses
 Suppose for two possible causes of a stiff neck, meningitis (m) and not
meningitis (¬m)
P ( Meningitis )    P ( s | m) P (m), P ( s | m) P (m) 
 We simply calculate the top line for each one and then normalise (divide
by the sum of the top line for all hypothesises
 But sometimes it’s harder to find out P(effect|cause) for all causes
independently than it is simply to find out P(effect)
 Note that Bayes’ rule here relies on the fact the effect must have arisen
because of one of the hypothesised causes. You can’t reason directly
about causes you haven’t imagined.
Bayes’ rule: combining evidence
 Suppose we have several pieces of evidence we want to
combine:
• John rings and Mary rings
• I have toothache and the dental probe catches on my tooth
 How do we do this?
P (cavity | toothache  catch)   P (toothache  catch | cavity) P (cavity)
 As we have more effects our causal model becomes very

complicated (for N binary effects there will be 2N different
combinations of evidence that we need to model given a
cause)
P(toothache  catch | cavity) , P(toothache  catch | cavity ) 

Bayes rule + conditional independence
 In many practical applications there are not a few evidence variables but
hundreds
 Thus 2N is very big
 This nearly led everyone to give up and rely on approximate or

qualitative methods for reasoning about uncertainty
 But conditional independence helps
 Toothache and catch are not independent, but they are independent
given the presence or absence of a cavity.
 In other words we can use the knowledge that cavities cause toothache
and they cause the catch, but the catch and the toothache do not cause
each other (they have a single common cause).
Bayes’ nets
 This can be captured in a picture, where the arcs capture conditional
independence relationships
 Or in a new equation:
P(cavity | toothache  catch)   P(toothache  catch | cavity) P(cavity)

  P (toothache | cavity ) P (catch | cavity ) P(cavity )
 Using conditional independence the causal model is much more

compact. Rather than the number of parameters being O(2 N) it is O(N)
where N is the number of effects (or evidence variables)

The Bayesian brain
 It turns out that your brain understands Bayes rule!
Reproduced without permission

from Kording & Wolpert
Nature Vol 427, 15 Jan 2004
 The human has to move their (hidden) hand to a target. Half

way through the movement they are given an uncertain clue
as to where the hand is. The position their hand moves to by
the end is consistent with the brain using Bayes’ rule to
combine the information it receives
The non-Bayesian brain
 But it is not apparent

that Bayes’ rule is used
everywhere
 How will rotation of one

wheel affect the other?
 Some kinds of
inference don’t seem to
be obviously
explainable using
probabilistic reasoning
alone

Summary
 Reasoning under uncertainty is an important area of AI
 It is not the case that statistical methods are the only way
 Logics can also cope with uncertainty in a qualitative way
 But statistical methods, and particularly Bayesian reasoning

has become a cornerstone of modern AI (rightly or wrongly)

Acting Under Uncertainty Basic Concepts

Uploaded by

Copyright:

Available Formats

Acting Under Uncertainty Basic Concepts

Uploaded by

Document Information

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

Acting Under Uncertainty Basic Concepts

Uploaded by

Copyright:

Available Formats

Plan

 Basic probability theory

 The Bayesian brain

 Reading: Russell &Norvig pp.462-488

AI Principles, Lecture on Reasoning Under Uncertainty

• When you travel you reason about the possibility of delays

 Systems which can reason about the effects of uncertainty

 But how should uncertainty be represented?

AI Principles, Lecture on Reasoning Under Uncertainty

 The domain of Throw is {head, tail}

 A Boolean random variable has two outcomes.

 Cavity has the domain {true, false}

 Toothache has the domain {true, false}

AI Principles, Lecture on Reasoning Under Uncertainty

 An atomic event is a complete specification of the values of the random

Toothache = true ^ Cavity = true

 The set of all possible atomic events has two properties:

AI Principles, Lecture on Reasoning Under Uncertainty

 Some simple rules governing probabilities

1. All probabilities are between 0 and 1 inclusive 0  P(a)  1

From these three laws all of probability theory can be derived.

AI Principles, Lecture on Reasoning Under Uncertainty

AI Principles, Lecture on Reasoning Under Uncertainty

 So in our toy example 1:

P (toothache  cavity)  P (toothache | cavity) P (cavity)

 But this doesn’t help us answer our question:

AI Principles, Lecture on Reasoning Under Uncertainty

 It is the cornerstone of modern probabilistic AI.

 But why is it useful?

AI Principles, Lecture on Reasoning Under Uncertainty

P (effect | cause) P (cause)

 In fact good models of P (effect | cause ) often available to us in real

AI Principles, Lecture on Reasoning Under Uncertainty

 0.05 in the population (1/50,000)

P ( s | m) P (m) 0.5  0.00002

 Because diagnostic knowledge is often more fragile than causal

 Suppose there was a meningitis epidemic? The rate of meningitis goes

P ( s | m) P (m) 0.5  0.0004

AI Principles, Lecture on Reasoning Under Uncertainty

 As we have more effects our causal model becomes very

AI Principles, Lecture on Reasoning Under Uncertainty

 Thus 2N is very big

 This nearly led everyone to give up and rely on approximate or

 But conditional independence helps

P(cavity | toothache  catch)   P(toothache  catch | cavity) P(cavity)

 Using conditional independence the causal model is much more

AI Principles, Lecture on Reasoning Under Uncertainty

Reproduced without permission

 The human has to move their (hidden) hand to a target. Half

 But it is not apparent

 How will rotation of one

AI Principles, Lecture on Reasoning Under Uncertainty

 Logics can also cope with uncertainty in a qualitative way

 But statistical methods, and particularly Bayesian reasoning

AI Principles, Lecture on Reasoning Under Uncertainty

You might also like

Pfad - The Proxy pFad of © 2024 Garber Painting. All rights reserved.