I wrote an article titled “Winnow Classification using C#” in the September 2014 issue of MSDN Magazine. See http://msdn.microsoft.com/en-us/magazine/dn781362.aspx. In machine learning, an example of a classification problem is to predict which class (male or female) a person belongs to, based on features such as occupation, political party, height and so on. There are many kinds of classification techniques, including neural network classifiers, logistic regression classification, and so on.
Winnow classification is a relatively little-used technique. Winnow classification applies when the dependent y-variable to predict is binary (two possible values, like sex) and also each of the independent predictor variables is also binary. The example I present in the MSDN article predicts whether a member of the U.S. Congress is a Democrat or a Republican, based on 16 predictor votes which can be “yes” or “no”.
In my opinion, there is no clear evidence that tells whether Winnow Classification is better, worse, or about equal to other classification techniques.