Tuesday, January 29, 2013

Phase Diagram and Approximate Message Passing for Blind Calibration and Dictionary Learning

Lenka Zdeborová sent me the following:

Dear Igor, 
Surely you will find this paper yourself as you always do ... but let me send a link anyway: http://arxiv.org/abs/1301.5898 . We think that this is a nice contribution to the matrix factorization jungle ...! For calibration (and others, completion etc.) the algorithm works for a number of samples just a bit larger than the trivial counting bound, which is much much lower than anything else we have seen. It needs some tuning to work really well in the dictionary learning case, but we think this is a very promising track.
At the same time, and again, if we missed some crucial references on the topic, please let us know.


We consider dictionary learning and blind calibration for signals and matrices created from a random ensemble. We study the mean-squared error in the limit of large signal dimension using the replica method and unveil the appearance of phase transitions delimiting impossible, possible-but-hard and possible inference regions. We also introduce an approximate message passing algorithm that asymptotically matches the theoretical performance, and show through numerical tests that it performs very well, for the calibration problem, for tractable system sizes.

Join the CompressiveSensing subreddit or the Google+ Community and post there !
Liked this entry ? subscribe to Nuit Blanche's feed, there's more where that came from. You can also subscribe to Nuit Blanche by Email, explore the Big Picture in Compressive Sensing or the Matrix Factorization Jungle and join the conversations on compressive sensing, advanced matrix factorization and calibration issues on Linkedin.

No comments: