Topics in Regularization and Boosting

  title={Topics in Regularization and Boosting},
  author={Saharon Rosset},
Regularization is critical for successful statistical modeling of “modern” data, which is high-dimensional, sometimes noisy and often contains a lot of irrelevant predictors. It exists — implicitly or explicitly — at the heart of all successful methods. The two main challenges which we take on in this thesis are understanding its various aspects better and suggesting new regularization approaches, which are motivated by both theoretical and computational considerations. This thesis comprises a… CONTINUE READING