[Colloquium] Amir Globerson: "Message Passing and Inference - A Dual Perspective" at TTI-C

[Gary Hamburg]

WHAT:

Amir Globerson: "Message Passing and Inference - A Dual Perspective"

 

WHEN:

Mon Nov 26 12:30pm - Mon Nov 26 1:30pm

 

WHERE:

TTI-C conference room

 

ABSSTRACT:

Graphical models are a powerful tool for representing distributions over
complex multivariate objects such as images or documents. Although graphical
models have been used with considerable success in many domains, such as
machine vision and signal processing, it is theoretically NP hard to infer
even simple model properties, such as the marginals over single variables,
or the most likely assignment. This difficulty has been addressed in
practice by designing approximate inference algorithms (such as belief
propagation) that often work well in practice, although with relatively weak
theoretical guarantees.

 

In this work we use the notion of convex duality to design a class of
message passing algorithms that solve convex variational formulations of
inference problems. Specifically, we use the geometric and linear
programming duals of two variational approximations, and show that
coordinate ascent on these duals yields a convergent message passing
algorithm.

 

We illustrate the performance of our new algorithms on various problems, and
show that they converge in cases where message passing algorithms such as
max product do not.

 

Host: Nati Srebro

 

Contact host at nati at tti-c.org to be placed on the speaker's schedule.

 

 

 

-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://mailman.cs.uchicago.edu/pipermail/colloquium/attachments/20071119/84fafc89/attachment-0001.html