Exact Calibration and Sharp Risk Geometry for Volume-Sampled Ridge Regression

Key points

• The determinant law and selected ridge fit share one positive definite penalty.1
• Our main result concerns centered covariance risk normalized by full-data penalized loss.1
• The balanced geometry yields a same-sample unbiased ridge--Horvitz--Thompson mixture with lower sharp risk and an exact mean-share improvement boundary.1
• Under full recalibration after feature changes, we prove quadratic regret from searching the complete old maximizing space and a query-uniform bound on the mixture's risk gain.1

Sources (1)

• [1]Exact Calibration and Sharp Risk Geometry for Volume-Sampled Ridge Regression
  arXiv (AI, ML, NLP, CV, robotics, multi-agent) · Oct 6, 04:16 AM

  We study ridge regression from exactly $s$ distinct rows of a fixed design.

  The determinant law and selected ridge fit share one positive definite penalty.

Extractive summary: sentences quoted from the sources.