A distributed block coordinate descent method for training $l_1$ regularized linear classifiers | Read Paper on Bytez