Friday, January 22, 2010

CS: Message passing on dense graphs with applications to compressed sensing , NIPS'09 videos


The dynamics of message passing on dense graphs, with applications to compressed sensing by Mohsen Bayati and Andrea Montanari. The abstract reads:
Approximate message passing' algorithms proved to be extremely effective in reconstructing sparse signals from a small number of incoherent linear measurements. Extensive numerical experiments further showed that their dynamics is accurately tracked by a simple one-dimensional iteration termed `state evolution'. In this paper we provide the first rigorous foundation to state evolution. We prove that indeed it holds asymptotically in the large system limit for sensing matrices with iid gaussian entries.
While our focus is on message passing algorithms for compressed sensing, the analysis extends beyond this setting, to a general class of algorithms on dense graphs. In this context, state evolution plays the role that density evolution has for sparse graphs.
also on the same subject, the 15th Annual LIDS student conference at MIT will feature Message-passing For Compressed Sensing by Venkat Chandar. The abstract of the talk is:

We propose a message-passing algorithm to recover a non-negative vector x from given linear measurements y=Ax, where A is an m-by-n matrix. The algorithm is very similar to the belief propagation algorithm(s) utilized in the context of decoding low density parity check codes. We establish that when A corresponds to the adjacency matrix of a bipartite graph with sufficient expansion, the algorithm produces a reconstruction r(x) of x satisfying

Formula,where x(k) is the best k-sparse approximation of x. The algorithm performs Formula computation in total, and the number of measurements required is Formula. In the special case when x is k-sparse, the algorithm recovers x exactly in time Formula. Conceptually, this work provides a rigorous connection between the theory of message-passing algorithms and compressed sensing that has been alluded to in many of the recent prior works.

Finally, the NIPS'09 videos are out. The ones that I mentioned before can be found below:
Of interest:
Thanks David Miraut Andres for the pointer.


Image Credit: NASA/JPL/Space Science Institute, Titan as seen by Cassini the day before yesterday on January 20th, 2010.

No comments:

Printfriendly