Skip to content
Open access

Transfer Learning Under Group‐Label Shift: A Semiparametric Exponential Tilting Approach

Oct 2026 · Scandinavian Journal of Statistics · 0 citations · 11 references

TL;DR

This article studies unsupervised transfer learning for a binary outcome, with labeled source data and unlabeled target covariates, and formulate this structure as group‐label shift (GLS) and pair it with a class‐specific exponential tilting model for the joint density ratio between the target and source populations.

Abstract

This article studies unsupervised transfer learning for a binary outcome, with labeled source data and unlabeled target covariates. The setting is motivated by cases where standard covariate‐shift and label‐shift assumptions are inadequate: group‐ or environment‐related features may vary jointly with the label across populations, while the remaining covariates remain conditionally stable given these features and the label. We formulate this structure as group‐label shift (GLS) and pair it with a class‐specific exponential tilting model for the joint density ratio between the target and source populations. The GLS framework generalizes existing shift assumptions by allowing heterogeneity in both feature and label distributions while mitigating spurious cross‐domain correlations. Under sufficient support and variation conditions on the source outcome regression, the tilting parameters are identifiable from the source distribution of covariates and outcomes, together with the target covariate distribution. A computationally efficient two‐step likelihood‐based estimation procedure is developed. We derive consistency and asymptotic normality for the resulting estimators and extend the theory to receiver operating characteristic curves, the area under the curve, and related target functionals, addressing the nonstandard challenges posed by plug‐in classifiers. Simulation studies demonstrate that the proposed method outperforms existing alternatives under GLS settings. An application to the Waterbirds data set further illustrates its effectiveness in transfer learning and classification.

Read PDF

Similar papers

Preprint Oct 2026

Beyond Source-Level Transfer: Sample-Level Learning for High-Dimensional Quantile Regression

This paper studies transfer learning for high-dimensional quantile regression with limited target samples and heterogeneous source domains. We propose a sample-level transfer learning method (SL-TL) that combines sample selection and importance weighting. Unlike existing source-level approaches that include or exclude...

Xiang-Yu Xing, Ying-Xuan Wang, Wang-Li Xu · 0 citations
Preprint Sep 2026

Grouped Semi-Supervised Estimation of Two-Sample Functionals via Single-Index Conditional Imputation

Estimating two-sample comparison functionals within heterogeneous groups is challenging when outcomes are observed only for a small labelled subset while covariates are widely available. We develop semi-supervised estimation of group-specific two-sample comparison functionals and their prespecified aggregates. The prop...

Heng Xu, T. Tan · 0 citations

Efficient Inference under Label Shift in Unsupervised Domain Adaptation

This paper considers a particular type of distribution shift, label shift, and develops an e-icient inference procedure for general parameters characterizing the unlabeled target population, and proposes a progressive estimation strategy that unfolds in three stages: an initial heuristic guess, a consistent estimation,...

Yan-Yuan Ma, Ji-Wei Zhao · 0 citations
Preprint Aug 2026

Separating Covariate Shift from Mechanism Change with Two Discriminators: CJSD, a Conditional Discrepancy with an Exact Covariate-Concept Decomposition

The Conditional Jensen-Shannon Discrepancy (CJSD) is proposed, which proves a covariate-null property, a drift-mass law, a one-sided misspecification-control inequality, a one-sided misspecification-control inequality, and a fixed-measure metrization via an identifiability lemma.

Kentaro Oda · 3 citations
Preprint Sep 2026

Reconciling Interpretability with Covariate-Dependent Shape Flexibility in Penalized Transformation Models for Distributional Regression

A central challenge in distributional regression is to allow the shape of the conditional distribution of the response variable to vary flexibly with covariates while retaining directly interpretable effects on its mean and standard deviation. We extend the penalized transformation model (PTM) family into a conditional...

Johannes Brachem, Thomas Kneib · 0 citations

We use cookies to run the site and, with your consent, for analytics and to show ads. See our Cookie Policy.