Summary of Methods
	• Simple: choose unlabeled instance closest to current SVM • Ratio: choose unlabeled instance for which m+ and m- are similar in size • Hybrid: do Ratio for first few queries then do Simple for the rest