The Optimal Ridge Penalty for Real-world High-dimensional Data Can Be Zero or Negative due to the Implicit Ridge Regularization

20 citations recorded by the source corpus.

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

Forward stability and model path selection (2021, neutral): “Indeed, our focus here is on the predictive accuracy of the models produced and some surprising recent research has suggested that noise features— completely independent of the response—may actually help improve model accuracy in some noisy data settings (Kobak et al. 2020; Mentch and Zhou 2022).”
Surprises in High-Dimensional Ridgeless Least Squares Interpolation (2019, background): “In contrast, [40] recently pointed out that —when β is aligned with the leading eigenvectors of Σ— min-norm regression can have optimal risk (i.”