Thursday, May 31, 2012

Faster Block Sparse Bayesian Learning Implementations

Zhilin sent me the following:

Dear Igor, 
Thank you for introducing my BSBL work in your blog. I have updated the BSBL codes 

The package includes BSBL-EM, BSBL-BO and EBSBL-BO. The codes are much faster than the old versions. Further, the package contains a fold, which has demo files to perform the whole experiment in my telemonitoring post i.e. using BSBL-BO to recover the fetal ECG recordings and then using FastICA to do decomposition. This may be convenient to those people who want to try the experiment but are not familiar with ICA. 

The posts  Zhilin  refers to are:

Thanks Zhilin

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.


TrekkingLemon said...

This looks interesting. I'd love to try it on regular ECG as well since they are not all that sparse. It might make a difference for our system of low power ECG compression using CS

Igor said...


For the regular ECG, you usually used MSE as a measure of "good" reconstruction ?

Keep us updated on the results of applying BSBL to your hardware!

Do you think BSBL could also be implmeneted on the iPhone ?


TrekkingLemon said...


we used MSE for the optimization but usually also report results based on metrics developed for ECG.
We will definitely try BBSL first to check if we can reduce the amount of measurements at the encoder since it does not have to be modified.
We'll have a look at implementing the decoder on the iPhone as well. Hard to say now if we can reach real-time performance but it is worth trying