Presentation is loading. Please wait.

Presentation is loading. Please wait.

Probabilistic Inference Reading: Chapter 13 Next time: How should we define artificial intelligence? Reading for next time (see Links, Reading for Retrospective.

Similar presentations


Presentation on theme: "Probabilistic Inference Reading: Chapter 13 Next time: How should we define artificial intelligence? Reading for next time (see Links, Reading for Retrospective."— Presentation transcript:

1 Probabilistic Inference Reading: Chapter 13 Next time: How should we define artificial intelligence? Reading for next time (see Links, Reading for Retrospective Class): Turing paper Mind, Brain and Behavior, John Searle Prepare discussion points by midnight, wed night (see end of slides)

2 2 Transition to empirical AI  Add in  Ability to infer new facts from old  Ability to generalize  Ability to learn based on past observation  Key:  Observation of the world  Best decision given what is known

3 3 Overview of Probabilistic Inference  Some terminology  Inference by enumeration  Bayesian Networks

4 4

5 5

6 6

7 7

8 8

9 9 Probability Basics  Sample space  Atomic event  Probability model  An event A

10 10

11 11 Random Variables  Random variable  Probability for a random variable

12 12

13 13

14 14

15 15

16 16

17 17 Logical Propositions and Probability  Proposition = event (set of sample points)  Given Boolean random variables A and B:  Event a = set of sample points where A(ω)=true  Event ⌐ a=set of sample points where A(ω)=false  Event aΛb=points where A(ω)=true and B(ω)=true  Often the sample space is the Cartesian product of the range of variables  Proposition=disjunction of atomic events in which it is true (aVb) = ( ⌐ aΛb)V(aΛ ⌐ b)V(aΛb) P(aVb)= P( ⌐ aΛb)+P(aΛ ⌐ b)+P(aΛb)

18 18

19 19

20 20

21 21

22 22

23 23

24 24

25 25 Axioms of Probability  All probabilities are between 0 and 1  Necessarily true propositions have probability 1. Necessarily false propositions have probability 0  The probability of a disjunction is  P(aVb)=P(a)+P(b)-P(aΛb)  P( ⌐ a)=1-p(a)

26 26  The definitions imply that certain logically related events must have related probabilities P(aVb)= P(a)+P(b)-P(aΛb)

27 27 Prior Probability  Prior or unconditional probabilities of propositions P(female=true)=.5 corresponds to belief prior to arrival of any new evidence  Probability distribution gives values for all possible assignments P(color) = (color = green, color=blue, color=purple) P(color)= (normalized: sums to 1)  Joint probability distribution for a set of r.v.s gives the probability of every atomic event on those r.v.s (i.e., every sample point) P(color,gender) = a 3X2 matrix

28 28

29 29

30 30

31 31

32 32

33 33

34 34 Inference by enumeration  Start with the joint distribution

35 35 Inference by enumeration  P(HasTeeth)= =.2

36 36 Inference by enumeration  P(HasTeethVColor=Green)= =.4 4

37 37 Conditional Probability  Conditional or posterior probabilities E.g., P(PlayerWins|HostOpenDoor=1 and PlayerPickDoor2 and Door1=goat) =.5 If we know more (e.g., HostOpenDoor=3 and door3-goat): P(PlayerWins)=1 Note: the less specific belief remains valid after more evidence arrives, but is not always useful  New evidence may be irrelevant, allowing simplification: P(PlayerWins|California- earthquake)=P(PlayerWins)=.3

38 38 Conditional Probability A general version holds for joint distributions: P(PlayerWins,HostOpensDoor1)=P(PlayerWins|HostOpensDoor1)*P(Ho stOpensDoor1)

39 39 Inference by enumeration  Compute conditional probabilities:  P( ⌐Hasteeth|color=green)= P(⌐HasteethΛcolor=green) P(color=green) 0.8 =

40 40 Normalization  Denominator can be viewed as normalization constraint α  P( ⌐Hasteeth|color=green ) = α P( ⌐Hasteeth|color=green ) =α[P( ⌐Hasteeth,color=green, female )+ P( ⌐Hasteeth,color=green, ⌐ female)] =α[ + ]=α =  Compute distribution on query variable by fixing evidence variables and summing over hidden variables

41 41 Inference by enumeration

42 42 Independence  A and B are independent iff P(A|B)=P(A) or P(B|A)=P(B) or P(A,B)=P(A)P(B)  32 entries reduced to 12; for n independent biased coins, 2 n -> n  Absolute independence powerful but rare  Any domain is large with hundreds of variables none of which are independent

43 43

44 44 Conditional Independence  If I have length <=.2, the probability that I am female doesn’t depend on whether or not I have teeth: P(female|length<=.2,hasteeth)=P(female|h asteeth)  The same independence holds if I am >.2  P(male|length>.2,hasteeth)=P(male|length>.2)  Gender is conditionally independent of hasteeth given length

45 45  In most cases, the use of conditional independence reduces the size of the representation of the joint distribution from exponential in n to linear in n  Conditional independence is our most basic and robust form of knowledge about uncertain environments

46 46 Next Class: Turing Paper  A discussion class  Graduate students and non-degree students: Anyone beyond a bachelor’s:  Prepare a short statement on the paper. Can be your reaction, your position, a place where you disagree, an explication of a point.  Undergraduates: Be prepared with questions for the graduate students  All: Submit your statement or your question by midnight Wed night.  All statements and questions will be printed and distributed in class on Wednesday.


Download ppt "Probabilistic Inference Reading: Chapter 13 Next time: How should we define artificial intelligence? Reading for next time (see Links, Reading for Retrospective."

Similar presentations


Ads by Google