Programming Techniques - fastICA ?

[censix]

I am currently looking for a code example or an idea on how to parallelize as much as possible of the ICA algorithm (Independent Component Analysis). ICA is an extension of Principal Component Analysis.

I have used PCA and ICA so far through high level software like R, which provides good sequential implementations in C for both.

What I am trying to see if a significant speedup can be achieved by making use of parallel processing and how this can be done, specifically with the E3/4 in mind.

Is anyone aware of existing solutions, (OpenCL ?)

regards

[censix]

Some google magic lead to this promising starting point

http://www.isip.uni-luebeck.de/index.php?id=479

unfortunately they use CUDA for the parallel ICA implementation, but openCL-ing it shouldn\t be a problem