Answers are grounded in this paper and cite their sources.
Demonstrates improved dimension reduction in limited labeled data environments, suggesting effective augmentation with unlabeled data.
Sliced inverse regression (SIR) is a popular framework for supervised dimension reduction, which projects high-dimensional covariates onto a low-dimensional linear subspace without information loss. In this article, we extend SIR to a semi-supervised setting with limited labeled data and abundant unlabeled data, and we explicitly exploit machine learning predictions to guide slice assignment. The proposed Prediction-Assisted Sliced Inverse Regression (PASIR) uses outcome predictions to determine whether unobserved responses fall into predefined slices, rather than imputing their exact values, and leverages the slice structure of SIR, which only depends on subgroup membership. We formulate the selection of unlabeled points as a convex quadratic program that jointly optimizes kernel-based diversity and slice-wise false selection rate (FSR) control under a fixed measurement budget for labels. Numerical simulations and a real-data example validate the superior performance of the proposed strategy.
No takes yet. Share an insight, caveat, or question.
Bao et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: