Seems to be a slightly modified Stochastic Gradient Descent to handle the data typically seen with this sort of analysis. https://github.com/JohnLangford/vowpal_wabbit/wiki/Learning-...