{[ promptMessage ]}

Bookmark it

{[ promptMessage ]}

# lec2 - CSci 5512 Artificial Intelligence II Instructor...

This preview shows pages 1–5. Sign up to view the full content.

This preview has intentionally blurred sections. Sign up to view the full version.

View Full Document

This preview has intentionally blurred sections. Sign up to view the full version.

View Full Document
This is the end of the preview. Sign up to access the rest of the document.

Unformatted text preview: CSci 5512: Artificial Intelligence II Instructor: Arindam Banerjee January 23, 2012 El Q I III M b p Q Instructor: Arindam Bancrycc Example Topology of network encodes conditional independence assertions Toothache @ 0 Weather is independent of the other variables 0 Toothache, Catch are conditionally independent given Cavity El Q I III M b p Q Instructor: Arindam Bancrycc — Bayesian networks A simple, graphical notation for conditional independence assertions and hence for compact specification of full joint distributions 0 Syntax o A set of nodes, one per variable a A directed, acyclic graph (link implies direct influence) 0 A conditional distribution for each node given its parents 0 Conditional distributions 0 For each X,-, P(X,-|Parents(X,-)) o In the form of a conditional probability table (CPT) 0 Distribution of X,- for each combination of parent values El 5' ' _= ORG“ Instructor: Arindam Bancrycc Example I 'm at work, neighbor John calls to say my alarm is ringing, but neighbor Mary doesn't call. Sometimes it's set off by minor earthquakes. Is there a burglar? 0 Variables: Burglar, Earthquake, Alarm, JohnCalls, MaryCalls 0 Network topology reflects ”causal" knowledge o A burglar can set the alarm off 0 An earthquake can set the alarm off a The alarm can cause Mary to call o The alarm can cause John to call El Q I III M b p Q Instructor: Arindam Bancrycc Example (Contd.) A _(J|A) @un @E- 0'0“ .0- El Q I III M b p Q Instructor: Arindam Bancrjcc Global semantics 0 Full joint distribution 0 Can be written as product of local conditionals 0 Example: PU, m,ambﬂe) = P(ﬁb)P(ﬁe)P(alﬁbaﬁe)PUl3)P(mla) 0 Example: PUﬁm, 3, baﬁe) = P(b)P(ne)P(alb,ﬁe)P(jl8)P(nMIa) 0 Can we compute P(b|j,-um)? El Q I III M b p Q Instructor: Arindam Bancrjcc ;@ @ﬁﬁi) o A CPT for Boolean X,- with k Boolean parents o 2" rows for the combinations of parent values 0 Each row requires one number 0 Each variable has no more than k parents 0 The complete network requires 0(n - 2") numbers - Grows linearly with n 0 Full joint distribution requires 0(2") 0 Example: Burglary network 0 Full joint distribution requires 25 — 1 = 31 numbers o Bayes net requires 10 numbers El .5: — —= Instructor: Arindam Bancrjcc MI 0 p Q Local semantics Each node is conditionally independent of its nondescendants given its parents El Q I III M b p Q Instructor: Arindam Bancrjcc Markov blanket Conditional Independence (Contd.) 0 o Gib/Q Q69 ®\'®/@ @ 9 (a) (b) (c) (d) Which BNs support x J. y|z o For (a)-(b), z is not a collider, so x J. y|z Each node is conditionally independent of all others given its Markov blanket, i.e., parents + children + children '5 parents 0 For (c), z is a collider, so x and y are conditionally dependent o For (d), w is a collider, and z is a descendent of w, so x and y are conditionally dependent El Q I III M b p ’9 El Q I III M b p Q Instructor: Arindam Bancrycc Instructor: Arindam Bancrycc Conditional Independence in EMS d—connection, d—separation @ a 6 ® a a e a 0 Definition (d-connection): X, Y,Z be disjoint sets of vertices in a directed graph G. X, Y is d—connected by Z iff El an a 9 9 @ undirected path U between some x e X,y E Y such that for every collider C on U (a) (b) (c) (d) 0 Either C or a descendent of C is in Z I No non—collider on U is in Z 0 Otherwise X and Y are d-separated by Z 0 If Z d-separates X and Y, then X J_ Y|Z for all distributions represented by the graph Which BNs support x1 J. X2|X3 o For (a), X1,X2 are dependent, X3 is a collider o For (b)-(d), X1 J. X2|X3 El Q I III M b p ’9 El Q I III M b p Q Instructor: Arindam Bancrycc Instructor: Arindam Bancrycc Conditional Independence (Contd.) Examples 0 For (a), a J. e o For (b) a and e are dependent given b; c and e are unconditionally dependent b; but a, e are dependent given {b, d} Instructor: Arindam Bancrycc Constructing Bayesian networks 0 Choose an ordering of variables X1, . . . ,X,, oFori=lton 0 Add X,- to the network 0 Select parents from X1, . . . 7X,-_1 such that P(X,-|Parents(X,-)) = P(XiIX1, - - - ,Xi—1) This choice of parents guarantees global semantics P(X1,...,X,,) n H P(X,-|X1, . . . ,X,-_1) i=1 H P(X,-|Parents(X,-)) i=1 El Q I III M b p Q Instructor: Arindam Bancrycc Conditional Independence: More Examples 0 (a) (b) o For (a), Is a J. c|e? Is a J. e|b? Is a J. e|c? o For (b), Is a J. e|d? Is a J. e|c? Is a J. c|b Instructor: Arindam Bancrycc Example: Car diagnosis Initial evidence: car won't start Testable variables (green), ”broken, so fix it" variables (orange) Hidden variables (gray) ensure sparse structure, reduce parameters Instructor: Arindam Bancrycc Example: Car insurance «w V-/.‘\ V @ \‘\\ 'mgﬂm EV ’V'z“\ m El Q I III M b p Q Instructor: Armdam Bancucc Inference F F .01 How can we compute P(bU,—um)? El Q I III M b p Q Instructor: Armdam Bancucc ...
View Full Document

{[ snackBarMessage ]}