University of Cambridge > Talks.cam > Computational and Biological Learning Seminar Series > Approximate Message Passing Algorithms

Approximate Message Passing Algorithms

Add to your list(s) Download to your calendar using vCal

If you have a question about this talk, please contact Zoubin Ghahramani.

Approximate Message Passing” (AMP) refers to a class of iterative algorithms that are Gaussian or quadratic approximations of loopy belief propagation algorithms on dense factor graphs. AMP has attracted widespread interest because it is significantly faster than traditional convex optimization procedures, particularly for solving the classic \ell_1-norm based optimization that is typical in sparse signal recovery (the LASSO ).

In the first part of the talk, I will discuss the main ideas behind AMP with examples. In the second part, I will describe an AMP decoding algorithm for sparse regression codes, a technique for communicating information over Gaussian noise channels. In this setting, the AMP decoder provably achieves the optimal information-theoretic limit (the channel capacity), and has excellent empirical performance as well. I will conclude with some open questions about AMP and its connections to classical optimization techniques.

This is joint work with Cynthia Rush and Adam Greig. The talk will be self-contained and will not assume prior knowledge of message passing/information theory/communications.

This talk is part of the Computational and Biological Learning Seminar Series series.

Tell a friend about this talk:

This talk is included in these lists:

Note that ex-directory lists are not shown.

 

© 2006-2017 Talks.cam, University of Cambridge. Contact Us | Help and Documentation | Privacy and Publicity