This action might not be possible to undo. Are you sure you want to continue?

Rochester, NY 14627, USA August 8, 2008

Reference: The material in this note is taken from Bishop, C. M. (2006). Pattern Recognition and Machine Learning. New York: Springer. Consider three variables a, b, and c, and suppose that the conditional distribution of a given b and c is such that it does not depend on the value of b [i.e., p(a|b, c) = p(a|c)]. In this case, a is conditionally independent of b given c. This can be expressed in a slightly diﬀerent way by considering the joint distribution of a and b given c, which can be written as follows: p(a, b|c) = p(a|b, c)p(b|c) = p(a|c)p(b|c) (1) (2)

Note that our deﬁnition of conditional independence requires that this equation holds for every possible value of c, and not just for some values. Consider the Bayesian network in Figure 1. Based on this graph, we can write the joint distribution of a and b given c as follows: p(a, b|c) = p(a, b, c) p(c) (3) (4)

= p(a|c)p(b|c)

meaning that a and b are conditionally independent given c. We can provide a simple graphical interpretation of this result by considering the path from a to b via c. Node c is said to be “tailto-tail” with respect to this path because the node is connected to the tails of the two arrows. When c is unobserved, the presence of this path causes a and b to be dependent. However, when c is observed and we condition on c, then the conditioned node “blocks” the path from a to b and causes a and b to become conditionally independent. Now consider the network in Figure 2. Based on this graph, the joint distribution of a and b given c is: p(a, b|c) = p(a, b, c) p(c) (5) (6) (7)

= p(a)p(c|a)p(b|c) = p(a|c)p(b|c)

1

again. In addition. this path is also blocked by e because e is a head-to-head node and neither it nor its descendants are in the conditioning set. conditioning on c unblocks the path and renders a and b dependent. the joint distribution of a and b given c is: p(a. the path from a to b is not blocked by f because it is a tail-to-tail node for this path and it is not observed. In the left graph. Then it can be shown that a head-to-head path becomes unblocked if either the node or any of its descendants is observed. then such a path connects a and b and renders them dependent. Consider a general directed graph in which A. meaning that a and b are conditionally independent given f . There is one more subtlety to this third example. First. people talk about the d-separation property for directed graphs. and a and b become conditionally independent. and C are arbitrary nonintersecting sets of nodes (whose union may be smaller than the complete set of nodes in the graph). we consider the case when node c is unobserved (left graph). a and b are conditionally independent given c. To determine this. Consider the graphs in Figure 4. and neither the node nor any of its descendants is in C. If all paths are blocked. a node y is a descendent of a node x if there is a path from x to y in which each step of the path follows the directions of the arrows. the path from a to b is blocked by node f because this is a tail-to-tail node that is observed. Based on this graph. However. Next. we consider the case when c is observed (right graph). we say that node c is “head-to-head” with respect to the path from a to b because it connects to the heads of the two arrows. a and b are not conditionally independent given c. nor is it blocked by node e because.meaning that. we consider all possible paths from any node in A to any node in B. b|c) = = p(a. Any such path is said to be blocked if it includes a node such that either: • the arrows on the path meet either head-to-tail or tail-to-tail at the node. consider the networks in Figure 3. In the right graph. Graphically. As a matter of terminology. and the variables a and b are independent. b) p(c) (8) (9) which does not factorize into the product p(a|c)p(b|c) meaning that a and b are not conditionally independent given c. b. it has a descendant c which is in the conditioning set. But if c is observed. B. then A is d-separated from B by C. If c is unobserved. We want to know if A and B are conditionally independent given C. c) p(c) p(a)p(b)p(c|a. 2 . In this case a and b are independent. In the graphical models literature. then this observation “blocks” the path. Lastly. although the latter is a head-to-head node. or • the arrows meet head-to-head at the node. The node c is said to be “head-to-tail” with respect to the path from a to b. it “blocks” the path. Thus. When c is unobserved. and the node is in the set C.

it is observed. 3 . a f e b a f e b c c Figure 4: Graphs referred to in text. The value of c is observed. a c b Figure 2: Node a is a parent of c which. in turn. the value of c is unobserved. a b a b c c Figure 3: Nodes a and b are parents of c. In the right graph. In the left graph. is a parent of b.c a b Figure 1: Node c is a parent of a and b. The value of c is observed.

- 127 Iio a New Subsystem
- Linux Cortexm Um 1.9.0 Ug (1)
- Tutorial DD
- lecture3_2
- La5
- In Tht Notes
- Proceedings ThermaComp2014 Website
- kheir
- Patankar-NHTFF-1980
- Transport
- Paul Summers Final Write Up
- Heat Conduction C Code Sample
- Fd Discret Ization
- kalman.c
- Computing Labs 2011
- 85275
- Le1999-PhDThesis
- c++ templates the complete guide
- frezza-buet13a
- Assignment 3 Kdtree
- Assignment 3 Kdtree
- c5734_a001
- Stnfrd50
- Contents

- Pnp
- Management Science Lecture Notes
- Graph Algorithms
- BC0052-Fall Drive Assignments 2012
- 4. Mathematics - IJMCAR - On Eccentric Connectivity Index - A. JAYENTHI
- Note
- 06 Graphs
- MIT14_15JF09_pajek
- Lecture 22
- Koofer(2)
- Bounds on Inverse Domination in Squares of Graphs
- Graphs
- Social Network Analysis
- Metaheuristic_NST_HCMC
- daa 6 & 7
- Strong (Weak) Triple Connected Domination Number of a Fuzzy Graph
- Split Block Domination in Graphs
- GetPDFServlet
- Tournaments Approximation Algorithms
- 09TSPbacktrack
- CM0167 Trees
- Theory of Computer Science
- BC0052 Theory of Computer Science
- Dynamic
- Lecture 1
- Trees With Unique Minimum
- DISTRIBUTION OF MAXIMAL CLIQUE SIZE OF THE VERTICES FOR THEORETICAL SMALL-WORLD NETWORKS AND REAL-WORLD NETWORKS
- Charaterization of Gaphs With Equal Domination and Connected Domination Numbers
- 2IL65 Assignment 7 Khanh
- Lecture_11_-_Graphs_P1_.pdf
- Conditional Ind

Are you sure?

This action might not be possible to undo. Are you sure you want to continue?

We've moved you to where you read on your other device.

Get the full title to continue

Get the full title to continue reading from where you left off, or restart the preview.

scribd