ASPICS: Applying Statistical Physics to Inference in Compressed Sensing

You probably recall ([1][2][3]) the paper entitled Statistical physics-based reconstruction in compressed sensing by Florent Krzakala, Marc Mézard, François Sausset,Yifan Sun, Lenka Zdeborová, a paper which promises to go beyond the Donoho-Tanner phase transition (provided some caveats).. Well  Florent Krzakala just sent me an email announcing the release of an implementation of their solver: ASPICS: Applying Statistical Physics to Inference in Compressed Sensing. A C++ and Python implementation of the solver is available here. 

For a short presentation, here are Marc Mezard's slides entitled Statistical physics approach to compressed sensing.

In the same way Jared Tanner's group provides a matlab file for the diverse Donoho-Tanner transitions, Florent's group also provides a similar text file for the Spinodal limit attained by the Belief Propagation algorithms. This limit is then passed over thanks to the use of the so-called seeded measurement matrixces explained in the paper.

Without the seeded matrices, the solver, seems to have similar performances as Jeremy Vila and Phil Schniter's EM-BG-AMP algorithm (see [3]) that I recently featured. Now that the codes are out, it shouldn't be hard to see how differently they behave, if they do.

I will add the code to the reconstruction section of the Big Picture in Compressive Sensing

Thanks Florent!

[1] A stunning development in breaking the Donoho-Tanner phase transition ?
[2] A short summary
[3] Build it and people will come

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.