We gratefully acknowledge support from
the Simons Foundation and member institutions.
Full-text links:

Download:

Current browse context:

cs.LG

Change to browse by:

References & Citations

DBLP - CS Bibliography

Bookmark

(what is this?)
CiteULike logo BibSonomy logo Mendeley logo del.icio.us logo Digg logo Reddit logo ScienceWISE logo

Computer Science > Machine Learning

Title: Near-Optimal Linear Regression under Distribution Shift

Abstract: Transfer learning is essential when sufficient data comes from the source domain, with scarce labeled data from the target domain. We develop estimators that achieve minimax linear risk for linear regression problems under distribution shift. Our algorithms cover different transfer learning settings including covariate shift and model shift. We also consider when data are generated from either linear or general nonlinear models. We show that linear minimax estimators are within an absolute constant of the minimax risk even among nonlinear estimators for various source/target distributions.
Comments: ICML 2021
Subjects: Machine Learning (cs.LG); Machine Learning (stat.ML)
Cite as: arXiv:2106.12108 [cs.LG]
  (or arXiv:2106.12108v1 [cs.LG] for this version)

Submission history

From: Qi Lei [view email]
[v1] Wed, 23 Jun 2021 00:52:50 GMT (4320kb,D)

Link back to: arXiv, form interface, contact.