Feature Bagging

Discriminative probabilistic models are very popular in NLP because of the latitude they afford in designing features. But training involves complex trade-offs among weights, which can be dangerous: a few …...