Andrew Frank 11022009 Joint work with Alex Ihler and Padhraic Smyth TexPoint fonts used in EMF Read the TexPoint manual before you delete this box A Graphical Models Nodes represent random variables ID: 783943
Download The PPT/PDF document "Belief Propagation in a Continuous World" is the property of its rightful owner. Permission is granted to download and print the materials on this web site for personal, non-commercial use only, and to display it on your personal computer provided you do not modify the materials and that you retain all copyright notices contained in the materials. By downloading content from our website, you accept the terms of this agreement.
Slide1
Belief Propagation in a Continuous World
Andrew Frank 11/02/2009Joint work with Alex Ihler and Padhraic Smyth
TexPoint fonts used in EMF.
Read the TexPoint manual before you delete this box.:
A
Slide2Graphical Models
Nodes represent random variables.Edges represent dependencies.
C
B
A
C
B
A
C
B
A
Slide3C
E
D
B
A
Markov Random Fields
E
D
B
C
A
D
A
C
E
B
B
E | C, D
A
C | B
Slide4Factoring Probability Distributions
Independence relations factorization
D
C
B
A
p(A,B,C,D) = f(A) f(B) f(C) f(D) f(A,B) f(B,C) f(B,D)
Slide5Toy Example: A Day in Court
W
A
E
V
A, E, W
є
{“Innocent”, “Guilty”}
V
є
{“Not guilty verdict”, “Guilty verdict”}
I
G
G
I
I
G
Slide6Inference
Most probable explanation:Marginalization:
Slide7Iterative Message Updates
x
Slide8Belief Propagation
W
A
E
V
m
AE
(E)
m
WE
(E)
m
EV
(V)
Slide9Loopy BP
C
A
B
D
C
A
B
D
Does this work? Does it make any sense?
Slide10A Variational Perspective
Reformulate the problem:
True distribution, P
“Tractable” distributions
Best tractable
approximation, Q
Find Q to minimize the
divergence
.
Slide11Desired traits:Simple enough to enable easy computation
Complex enough to represent P Choose an Approximating Family
e.g.
Fully factored:
Structured:
Slide12Choose a Divergence Measure
Kullback-Liebler divergence:
Alpha divergence:
Common choices:
Slide13Behavior of α-Divergence
Source: T.
Minka
. Divergence measures and message passing.
Technical Report MSR-TR-2005-173, Microsoft. Research, 2005.
Slide14Resulting Algorithms
Assuming a fully-factored form of Q, we get…*Mean field, α = 0Belief propagation, α = 1Tree-reweighted BP, α ≥ 1
* By minimizing “local divergence”:
Q(X
1
, X
2
, …,
Xn) = f(X1) f(X2) … f(X
n)
Slide15Local vs. Global Minimization
Source: T. Minka. Divergence measures and message passing. Technical Report MSR-TR-2005-173, Microsoft. Research, 2005.
Slide16Applications
Slide17Sensor Localization
A
B
C
Slide18Protein Side Chain Placement
RTDCYGN
+
Slide19Common traits?
?
Continuous state space:
Slide20Easy Solution: Discretize!
10 bins
10 bins
Domain size:
d = 100
20 bins
20 bins
Domain size:
d = 400
Each message:
O(d
2
)
Slide21Particle BP
We’d like to pass “continuous messages”…
C
A
B
D
B
m
AB
(B)
1
4
4.2
5
2.5
…
…
…
Instead, pass discrete messages over sets of particles:
{ b
(
i
)
} ~ W
B
(B)
m
AB
({b
(
i
)
})
b
(1)
b
(2)
b
(N)
. . .
Slide22PBP: Computing the Messages
Re-write as an expectation:
Finite-sample approximation:
Slide23Choosing“Good” Proposals
C
A
B
D
Proposal should “match” the integrand.
Sample from the belief:
Slide24Iteratively Refine Particle Sets
(2)
f(
x
s
,
x
t)
Draw a set of particles, {xs(i)} ~ Ws(
xs).Discrete inference over the particle discretization.
Adjust Ws(xs)
(1)
(3)
XsX
t(1)
(3)
Slide25Benefits of PBP
No distributional assumptions.Easy accuracy/speed trade-off.Relies on an “embedded” discrete algorithm.
Belief propagation, mean field, tree-reweighted BP…
Slide26Exploring PBP: A Simple Example
x
s
||
x
s
–
x
t
||
Slide27Continuous Ising Model
Marginals
Approximate
Exact
Mean Field PBP
α = 0
PBP
α = 1
TRW PBP
α = 1.5
* Run with 100 particles per node
Slide28A Localization Scenario
Slide29Exact Marginal
Slide30PBP Marginal
Slide31Tree-reweighted PBP Marginal
Slide32Estimating the Partition Function
Mean field provides a lower bound.Tree-reweighted BP provides an upper bound.p(A,B,C,D) = f(A) f(B) f(C) f(D) f(A,B) f(B,C) f(B,D)
Z = f(A) f(B) f(C) f(D) f(A,B) f(B,C) f(B,D)
Slide33Partition Function Bounds
Slide34Conclusions
BP and related algorithms are useful!Particle BP let’s you handle continuous RVs.Extensions to BP can work with PBP, too.Thank You!