Complexity regularization using data-dependent penalties

摘要

We define a regression function estimate based on complexityregularization, where the list of candidate functions and thecorresponding penalties are determined from the training data, leadingto improved performance