Transfer learning for piecewise-constant mean estimation: Optimality, l₁- and l₀-penalization

Summary

We study transfer learning for estimating piecewise-constant signals when source data, which may be relevant but disparate, are available in addition to the target data. We first investigate transfer learning estimators that respectively employ l₁- and l₀-penalties for unisource data scenarios and then generalize these estimators to accommodate multisources. To further reduce estimation errors, especially when some sources significantly differ from the target, we introduce an informative source selection algorithm. We then examine these estimators with multisource selection and establish their minimax optimality. Unlike the common narrative in the transfer learning literature that the performance is enhanced through large source sample sizes, our approaches leverage higher observational frequencies and accommodate diverse frequencies across multiple sources. Our extensive numerical experiments show that the proposed transfer learning estimators significantly improve the estimation performance compared to the estimators only using the target data.

Accepted manuscripts

Accepted manuscripts are PDF versions of the author’s final manuscript, as accepted for publication by the journal but prior to copyediting or typesetting. They can be cited using the author(s), article title, journal title, year of online publication, and DOI. They will be replaced by the final typeset articles, which may therefore contain changes. The DOI will remain the same throughout.