flâneur

13.pdf

people.eecs.berkeley.edu · 1,765 words · saved by 1 readers

N/A

Shrinkage: Ridge Regression, Subset Selection, and Lasso 71 13 Shrinkage: Ridge Regression, Subset Selection, and Lasso RIDGE REGRESSION aka Tikhonov Regularization Least-squares linear regression + `2 penalized mean loss. (1) + (A) + (a) + (d). Find w that minimizes kXw yk2 + kw0 k2 = J(w) where w0 is w with component ↵ replaced by 0. X has fictitious dimension but we DON’T penalize ↵. Adds a regularization term, aka a penalty term, for shrinkage: to encourage small kw0 k. Why? 1. Guarantees positive definite normal eq’ns;…

related reading