Multi-Source Transfer Learning of Sparse Single-Index Models

Ye Tian · 2026

Read Multi-Source Transfer Learning of Sparse Single-Index Models by Ye Tian (2026) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2606-29658.txt.

Description

Tian proposes a transfer learning framework that borrows strength from related datasets without ever seeing their raw data. Transfer learning improves a target task using knowledge from source domains. Existing theory mostly assumes a known link function, or some other linear structure, and requires direct access to source data.

His method rests on the single-index model, where outcomes depend on one combined score passed through an unknown curve. Only summary statistics cross from each source, derived through a generalized Stein's lemma, in a single one-time communication. That design preserves privacy and avoids harm from differing unknown link functions across domains. A multilayer perceptron, guided by the index estimated from the transferred statistics, captures the flexible nonlinearity and limits overfitting. Experiments on synthetic and real-world data show consistent improvements over generalized linear model based approaches.

Catalog record

CREATORYe Tian
YEAR2026
COLLECTIONPAPERS
FORMATTXT
FILE SIZE78 KB
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2606.29658

Browse the full library · About the Tower of Babel