Voxel selection using l1-norm

Post date: Aug 15, 2012 7:49:13 PM

L1-norm sends some coefficients to zero right away, so we don’t need to set a threshold for the coefficients like in l2-norm case where coefs are small but not zero.

Therefore, l1-can be used as a feature selection unit and usually gives better features than the traditional feature selection frameworks like MI as l1-norm consider the whole feature simultaneously.

In fact, someone have recently done it before using l1- and l2-norm on logistic regression:

http://www.ncbi.nlm.nih.gov/pmc/articles/PMC2856747/

So, I might want to read the paper and see how can we can make a difference.