You are developing a deep learning model and have training data with a large number of features. You are not sure which features are important. You'd like to use a regularization technique that will drive the parameter for the least important features toward zero. What regularization technique would you use?