The quest for the fastest SVM learning algorithm is continuing.

Leon Bottou reported his suprising finding: a classic optimization method, Stochastic Gradient Descent, is amazingly fast for training linear SVMs or CRFs. His program svmsgd works much faster than