数据芳榳论坛 | Multi-Source Prediction-Powered Inference 

报告题目：Multi-Source Prediction-Powered Inference 

主讲嘉宾：张新雨 研究员

讲座时间：2026年11月10日上午10:00-11:30

地点：博学楼三层阶7

主办单位：统计与数据科学学院

摘要：Prediction-powered inference integrates a small gold-standard dataset with large pseudo-labeled data generated by machine learning methods to enhance statistical inference. In modern applications, multiple data sources and diverse machine learning methods often give rise to multiple pseudo-labeled datasets, each encoding potentially different aspects of the underlying information. However, how to optimally combine multiple data sources and machine learning methods for statistical inference remains unclear. To address this problem, we propose a multi-source prediction-powered inference method by aggregating multiple pseudo-labeled datasets, with aggregation weights estimated by minimizing the estimated confidence-region volume. We study homogeneous settings, where source and target distributions coincide, and heterogeneous settings, where distributional discrepancies arise between source and target distributions, including covariate shift and domain shift. Theoretically, we establish asymptotic normality of the proposed estimator and asymptotic volume optimality within the prespecified weighting class. We further characterize when our method yields smaller confidence regions than classical target-only inference and single-source prediction-powered inference. Simulation studies show that our method can reduce confidence-region volume while maintaining inferential validity in the settings considered. An application to high body fat prevalence illustrates the reductions in estimated variance achieved by the proposed method. 

嘉宾简介：张新雨，中国科学院数学与系统科学研究院研究员。长期从事统计和计量经济学理论与应用方面的研究工作，与合作者解决了模型平均研究中的多个难题，并将模型平均与迁移学习、随机森林、投资组合、库存管理等融合提出了新方法，同时将所提出的预测方法应用于实际问题为相关部门的决策提供了参考依据。先后主持国家自然科学基金委青年基金A类及其延续项目等。

欢迎我校师生参加。