Tuesday, August 11, 2009

CS: Update on "A Simple Compressive Sensing Algorithm for Parallel Many-Core Architectures", Saturn Equinox

Jerome Darbon just let me know the following
We have just updated our report "A Simple Compressive Sensing Algorithm for Parallel Many-Core Architectures", Alexandre Borghi, Jerome Darbon, Sylvain Peyronnet, Tony Chan and Stanley Osher. We have added more experiments including time results for the newest Nvidia GPUs and multi-cores of Intel. You might be interested by these new results.

The new version is available here:
or as a webpage here:

This is very interesting as we now have information about what happens beyond some toy models. The first version of this paper was covered earlier when it came out. Let us note that other authors and teams have since used GPUs to perform the compressive sensing reconstruction (see the second part of the Compressive Sensing Hardware page). From that list, there are four:

2.1 UCLA GPU/Multicore solver.

2.2 University of Calgary GPU solver.

2.3 Graz University of Technology GPU solver

2.4 University of Wisconsin Implementation of SpaRSA on a GPU

Let us note that the SpaRSA paper is specifically designed around implementing non greedy solver on the GPU. One should also keep in mind, as I mentioned earlier this year, that there is always the Jacket product allowing Matlab to GPU programming. The paper of Jerome and co-workers covers multicore CPUs, an area of investigation left untouched by the other investigations, even though it looks like this type of approach is bound for a bright future. Thanks Jerome for the heads-up!

By the way, today is Saturn Equinox, an event that takes place every fifteen years, which means two things:

Image Credit: NASA/JPL/Space Science Institute.

No comments: