Winnow Algorithm

Winnow Algorithm is a for learning a linear classifier/decision hyper-plane from labeled examples. It scales well to high dimensions especially when many of the dimensions are irrelevant. Thus it finds good use for text classification problems using a bag-of-words feature representation.

Relevant Papers

1. Nick Littlestone (1988). "Learning Quickly When Irrelevant Attributes Abound: A New Linear-threshold Algorithm", Machine Learning
2. [Littlestone, Nicholas. "Mistake bounds and logarithmic linear-threshold learning algorithms." (1990).]