how to perform regularization in machine learning