What is the difference between Generative and Discriminative Learning?

Generally speaking in generative learning one is trying to optimize the parameters of a generative model to best explain the data that is observed. In discriminative learning there is no model assumed, but instead a function is optimized to best discriminate among classes. In the second case no assumption is made about the underlying process which is generating the data.

Generative	Discriminative
Maximize P(C,A1,...,AN)	Maximize P(C | A1,...,AN)
Example: Naive Bayes	Example: Logistic Regression
	A Bayes Net with a restricted CPT
Learning is just a matter of counting the training data.	Learning takes more work and is an optimization procedure