Professional Documents
Culture Documents
Dimitris Margaritis
(joint work with Facundo Bromberg and Vasant Honavar)
Department of Computer Science Iowa State University
The problem
General problem:
Specific problem:
2 / 66
can use them to calculate the probability of any propositional formula (probabilistic inference) given the facts (known values of some variables)
Efficient representation of the joint probability using conditional independences Most popular graphical models:
3 / 66
Markov Networks
Define neighborhood structure among variables (i, j):
Intuitively: variable X is conditionally independent (CI) of variable Y given set of variables Z if Z shields any influence between X to Y
5 / 66
6 / 66
V: nodes represent random variables, E: undirected edges represent structure i.e., (i; j ) 2 E () (i; j ) 2 N
N = f(1; 4); (4; 7); (7; 0); (7; 5); (6; 5); (0; 3); (5; 3); (3; 2)g
Example MN for:
V = f0,1,2,3,4,5,6,7g
7 / 66
? 3? = 7 j f0g ? 7 j f0; 5g 3?
(Pearl 88) If the CIs in the graph match exactly those of distribution P, P is said to be graph-isomorph.
8 / 66
Pr(1,2,
; 7)
True network
Learned network
9 / 66
Independence based
Infer graph using information of independences that hold in underlying model
10 / 66
Independence-based approach
Assumes existence of independence-query oracle that answers the CIs that hold in the true probability distribution Proceeds iteratively:
1. Query independence query oracle for CI value h in true model
2. Discard structures that violate CI h 3. Repeat until a single structure is left (uniqueness under assumptions)
Is variable 7 independent of variable 3 given variables {0,5}?
Can be approximated by a statistical independence test (SIT) e.g. Pearsons c2 or Wilks G2 Given as input:
a data set D (sampled from the true distribution), and a triplet (X,Y | Z)
The SIT computes the p-value: probability of error in assuming dependence when in fact variables are independent and decides:
12 / 66
Outline
Introductory Remarks The GSMN and GSIMN algorithms The Argumentative Independence Test Conclusions
13 / 66
14 / 66
GSMN algorithm
We introduce (the first) two independence-based algorithms for MN structure learning: GSMN and GSIMN GSMN (Grow-Shrink Markov Network structure inference algorithm) is a direct adaptation of the growshrink (GS) algorithm (Margaritis, 2000) for learning a variables Markov blanket using independence tests
Denition: A Markov blanket B L(X ) of X 2 V is any subset S of variables ? V S fX g j S). that shield X from all others variables, that is, (X ?
15 / 66
GSMN (contd)
N Markov blanket is the set of neighbors in the structure (Pearl and Paz 85). Therefore, we can learn the structure by learning the Markov blankets:
1: 2: 3: 4: for every X 2 V BL(X ) get Markov blanket of X using GS algorithm. for every Y 2 BL(X ) add edge (X; Y ) to E(G):
GSMN extends above algorithm with heuristic ordering for grow and shrink phases of GS
16 / 66
Initially No Arcs
F G
A D
17 / 66
Growing phase
2. F dependent of A given {B}? 1. B dependent of A given {}?
F G
Shrinking phase
Minimum Markov Blanket
F G
A D
GSIMN
GSIMN (Grow-Shrink Inference Markov Network) uses properties of CIs as inference rules to infer novel tests, avoiding costly SITs. Pearl (88) introduced properties satisfied by the CIs of distributions isomorphic to Markov networks:
66
21 / 66
Triangle theorems
GSIMN actually uses the Triangle Theorem rules, derived from (only): Strong Union and Transitivity:
? W j Z ) ^ (W 6? ?Y jZ ) (X 6? 1 2 6?Y jZ \Z ) =) (X ? 1 2
? W j Z ) ^ (W 6? ?Y jZ [Z ) (X ? 1 1 2 ? Y j Z ): =) (X ? 1
Rearranges GSMN visit order to maximize benefits Applies these rules only once (as opposed to computing the closure) Despite these simplifications, GSIMN infers >95% of 22 / inferable tests (shown experimentally)
66
Experiments
Our goal: Demonstrate GSIMN requires fewer tests than GSMN, without significantly affecting accuracy
23 / 66
24 / 66
25 / 66
26 / 66
Real-world data
Non-random topologies (e.g. regular lattices, small world, chains, etc.) Underlying distribution may not be graph-isomorph
27 / 66
Outline
Introductory Remarks The GSMN and GSIMN algorithms The Argumentative Independence Test Conclusions
28 / 66
29 / 66
The Problem
Statistical Independence tests (SITs) unreliable for small data sets Produce erroneous networks when used by independence-based algorithms This problem is one of the most important criticisms of independence-based approach
Our contribution
A new general purpose independence test: the argumentative independence test or AIT that improves reliability for small data sets
30 / 66
Main Idea
The new independence test (AIT) improves accuracy by correcting outcomes of a statistical independence test (SIT):
Incorrect SITs may produce CIs inconsistent with Pearls properties of conditional independences Thus, resolving inconsistencies among SITs may correct the errors propositions are CIs (i.e., for (X, Y | Z), or ) inference rules are Pearls conditional independence axioms
31 / 66
Pearls axioms
We presented above the undirected axioms Pearl (1988) also introduced, for any distribution:
general axioms
Directed axioms
32 / 66
Example
Consider the following KB of CIs, constructed using a SIT. ? 1 j f2 ; 3 g) (0 ? A. ? 4 j f2 ; 3 g) (0 ? B. (0 6? ? f1 ; 4 g j f2; 3g ) C. Assume C is wrong (SITs mistake). Assuming the Composition axiom holds, then D.
? 1 j f2; 3g ) ^ ( 0 ? ? 4 j f2; 3g) =) (0 ? ? f 1; 4g j f 2; 3g ) (0 ?
Example (contd)
At least two ways to resolve inconsistency: rejecting D or rejecting C If we can resolve inconsistency in favor of D, error could be corrected The argumentation framework presented next provides a principled approach for resolving inconsistencies
Consistent but and Inconsistent Consistent correct KB: Incorrect Incorrect KB:
D.
A. B. C.
? 1 j f2 ; 3 g) (0 ? ? 4 j f2 ; 3 g) (0 ? ? f 1 ; 4 g j f 2; 3g ) (0 6?
? 1 j f 2; 3g ) ^ ( 0 ? ? 4 j f2; 3g) =) (0 ? ? f 1; 4g j f 2; 3g ) (0 ?
34 / 66
Arguments
In independence KBs if-then rules are instances (propositionalizations) of Pearls universally quantified rules. For example these
are instances of Weak Union: Propositional arguments: arguments ({h}, h) for individual CI proposition h
36 / 66
Example
The set of arguments corresponding to KB of previous example is:
Name H, ? h) ? 1 j f2; 3g)g;(( ( f (0 ? 0 ? 1 j f2; 3g) ) ? 4 j f 2 ; 3 g ) g ; (0 ? ? 4 j f2; 3g) ) ( f (0 ? ? f1; 4g j f2; 3g)g; (0 6? ? f 1 ; 4 g j f 2 ; 3 g )) (f(0 6?
f( 0 ? ? 1 j f2 ; 3 g) ; ( 0 ? ? 4 j f 2 ; 3 g ) g ; (0 ? ? f1 ; 4 g j f 2 ; 3 g)
Correct?
A. B. C. D.
37 / 66
Preferences
Preference over arguments obtained from preferences over CI propositions We say argument (H, h) preferred over argument (H, h) iff it is more likely for all propositions in H to be correct:
The probability n(h) that h is correct is obtained from p-value of h, computed using a statistical test (SIT) on data
38 / 66
Example
Lets extend the arguments with preferences:
Name
A. B. C. D.
f( 0 ? ? 1 j f2 ; 3 g) ; ( 0 ? ? 4 j f 2 ; 3 g ) g ; (0 ? ? f1 ; 4g j f 2; 3g)
Correct?
n(H)
39 / 66
Attack relation
The attack relation formalizes and extends the notion of logical contradiction:
Definition: Argument b attacks argument a iff b logically
contradicts a and a is not preferred over b Since argument (H1,h1) models if H then h rules, it can be logically contradicted by (H2,h2) if:
(H1,h1) rebuts (H2,h2) iff h1 h2 (H1,h1) undercuts (H2,h2) iff $hH2 such that h h1
40 / 66
Example
Name
A. B. C. D.
Correct?
n(H)
41 / 66
Inference = Acceptability
42 / 66
Example
Name
A. B. C. D.
Correct?
n(H)
We had that D attacks C (and no other attack). Since nothing attacks D, D is accepted. C is attacked by an accepted argument, so C is rejected. Argumentation resolved the inconsistency in favor of correct proposition D! In practice, we have thousands of arguments. How to compute acceptability status of all of them?
43 / 66
44 / 66
45 / 66
46 / 66
47 / 66
48 / 66
Top-down algorithm
accept if all attackers are rejected, and reject if at least one attacker is accepted
49 / 66
Target node
1 2 6 3 4 8 13 5 7 10 11 9 12
50 / 66
Target node
1 2 6 3 4 8 13 5 7 10 11 9 12
11
attackers
51 / 66
Target node
1 2 6 3 4 8 13 5 7 10 11 4 5 12 9 12
11
attackers
52 / 66
Target node
1 2 6 3 4 8 13 5 7 10 11 4 5 12 9 12
11
leaf
53 / 66
Target node
1 2 6 3 4 8 13 5 7 10 2 1 11 4 5 12 9 12
11
leaf
13
leaf
leaf
leaf
54 / 66
Target node
1 2 6 3 4 8 13 5 7 10 2 1 13 11 4 5 12 9 12
11
55 / 66
Target node
1 2 6 3 4 8 13 5 7 10 2 1 13 11 4 5 12 9 12
11
56 / 66
Target node
1 2 6 3 4 8 13 5 7 10 2 1 13 11 4 5 12 9 12
11
57 / 66
Time complexity:
O(bd)
d=3
Difficulties: 1. Exponential in depth d. 2. By nature of Pearl rules, # attackers of some nodes (branching factor b) may be exponential Approximation: To solve (1), we limit d to 3. To solve (2), we consider an alternative propositionalization of Pearls rules that bounds b to polynomial size (details 58 / 66 omitted here)
Experiments
We considered 3 variations of each AIT, one per set of Pearl axioms: general, directed, and undirected Experiments on data sampled from Markov and Bayesian networks (directed graphical models)
59 / 66
PC algorithm
GSMN algorithm
61 / 66
62 / 66
Conclusions
63 / 66
Summary
64 / 66
Further Research
Use such constraints to improve Bayesian network and Markov network structure learning from small data sets (instead of just improving individual tests) Develop faster methods of inferring independences using Pearls axiomsProlog tricks?
65 / 66
66 / 66