Statistical estimation often involves tradeoffs between expensive, high-quality measurements and a variety of lower-quality proxies. We introduce Multiple-Prediction-Powered Inference (MultiPPI): a general framework for constructing statistically efficient estimates by optimally allocating resources across these diverse data sources. This work provides theoretical guarantees about the minimax optimality, finite-sample performance, and asymptotic normality of the MultiPPI estimator. Through experiments across three diverse large language model (LLM) evaluation scenarios, we show that MultiPPI consistently achieves lower estimation error than existing baselines. This advantage stems from its budget-adaptive allocation strategy, which strategically combines subsets of models by learning their complex cost and correlation structures.
Prediction-powered Inference by Mixture of Experts
Department of Mathematics, Hong Kong University of Science and Technology(香港理工大学数学系);Department of Mathematical and Statistical Sciences, University of Alberta(阿尔伯塔大学数学与统计学系)
M-estimation under Two-Phase Multiwave Sampling with Applications to Prediction-Powered Inference
Institute for Data, Systems, and Society, Massachusetts Institute of Technology(麻省理工学院数据、系统与社会研究所);Department of Electrical Engineering and Computer Science, Massachusetts Institute of Technology(麻省理工学院电气与计算机工程系)
Prediction-Powered Inference with Inverse Probability Weighting
Department of Statistics, Virginia Tech, Blacksburg, VA.(统计系,弗吉尼亚理工大学,弗吉尼亚州黑斯堡);Booth School of Business, University of Chicago, Chicago, IL.(商学院,芝加哥大学,伊利诺伊州芝加哥)
Center for Forecasting Science, Academy of Mathematics and Systems Science, Chinese Academy of Sciences(中国科学院数学与系统科学研究院预测科学中心);State Key Laboratory of Mathematical Sciences, Academy of Mathematics and Systems Science, Chinese Academy of Sciences(中国科学院数学与系统科学研究院数学科学学院国家重点实验室);School of Management, University of Science and Technology of China(中国科学技术大学管理学院)
Recidivism Prediction, Peer Effect Estimation, and Prediction-Powered Inference with LLM Text Measures
Department of Agricultural and Applied Economics, University of Georgia(农业与应用经济学系,佐治亚大学);Department of Statistics, The Ohio State University(统计系,俄亥俄州立大学);College of Social Work, The Ohio State University(社会工作学院,俄亥俄州立大学)
Department of Electrical and Computer Engineering, Technion IIT, Israel(电气与计算机工程系,技术离子研究所,以色列);Department of Statistics and Data Science, The Wharton School, University of Pennsylvania, USA(统计学与数据科学系,沃顿商学院,宾夕法尼亚大学,美国);Department of Computer Science, Technion IIT, Israel(计算机科学系,技术离子研究所,以色列)
Department of Electrical and Computer Engineering, Technion IIT, Israel(技术学院电气与计算机工程系);Department of Statistics and Data Science, The Wharton School, University of Pennsylvania, USA(宾夕法尼亚大学沃顿商学院统计与数据科学系);Department of Computer Science, Technion IIT, Israel(技术学院计算机科学系)
Department of Biostatistics, Columbia Mailman School of Public Health(哥伦比亚大学生物统计学系、哥伦比亚大学马利曼公共卫生学院);Institute for Data, Systems, and Society, Massachusetts Institute of Technology(麻省理工学院数据、系统和社会研究所);Department of Biostatistics, Yale School of Public Health(耶鲁大学生物统计学系)
Prediction-Powered Inference with Imputed Covariates and Nonuniform Sampling
Institute for Data, Systems, and Society, Massachusetts Institute of Technology(数据、系统与社会研究所,麻省理工学院);Department of Electrical Engineering and Computer Science, Massachusetts Institute of Technology(电气工程与计算机科学系,麻省理工学院);Stanford Data Science and Department of Statistics, Stanford University(斯坦福大学数据科学与统计学系)
CheckSupport: A Local LLM-Powered Tool for Automated Manuscript Submission Checklist Selection and Completion
Department of Radiology, Perelman School of Medicine at University of Pennsylvania(宾夕法尼亚大学佩雷尔曼医学学院放射科);Department of Computer Science, Drexel University(德雷塞尔大学计算机科学系);Department of Computer and Information Science, School of Engineering and Applied Science, University of Pennsylvania(宾夕法尼亚大学工程与应用科学学院计算机与信息科学系);Department of Radiology, Cooper University Hospital(科珀大学医院放射科);University of California Davis Graduate School of Management(加州大学戴维斯分校管理研究生院)
Synthetic-Powered Multiple Testing with FDR Control
Department of Statistics and Data Science, The Wharton School, University of Pennsylvania, USA(统计与数据科学系,沃顿商学院,宾夕法尼亚大学);Department of Electrical and Computer Engineering, Technion IIT, Israel(电气工程与计算机科学系,技术学院(IIT),以色列);Department of Computer Science, Technion IIT, Israel(计算机科学系,技术学院(IIT),以色列)
Prediction-Powered Linear Regression: A Balance Between Interpretation and Prediction
School of Management, University of Science and Technology of China(中国科学技术大学管理学院);School of Mathematics and Statistics, Jiangsu Normal University(江苏师范大学数学与统计学院);Academy of Mathematics and Systems Science, Chinese Academy of Sciences(中国科学院数学与系统科学研究院)