0.25
0.5
0.75
1.25
1.5
1.75
2
Approximations with Reweighted Generalized Belief Propagation
Published on Feb 25, 20073527 Views
In (Wainwright et al., 2002) a new general class of upper bounds on the log partition function of arbitrary undirected graphical models has been developed. This bound is constructed by taking convex c
Chapter list
Approximation with Reweighted Generalized Belief Propagation00:02
Content00:57
Motivation: Approximate interference with GBP01:57
Aproximations03:20
Exact model, partition function and free energy05:04
Junction trees06:14
Upper bound of log Z (Wainwright et al)09:29
Upper bound using junctions trees10:48
Convexified Kikuchi free energy12:14
Convexified Kikuchi free energy (cont.)15:27
Relation with Kikuchi free energy17:02
Relation with variational mean field free energy20:10
Reweighted Generaliized Belief Propagation (RGBP)20:55
RGBP(2): Propagation = reparametrization21:01
Counting numbers in some regular Ising models21:03
Fully connected Ising models24:10
Experimental set up24:47
Exerimental resaults: Torus26:41
Exerimental resaults: Fully connected model30:19
Consistency32:46
Summary, Conclusi34:58