New statistical methods boost efficiency using large auxiliary datasets

This paper explores methods for enhancing statistical efficiency by combining a small target dataset with a large auxiliary dataset. Researchers investigate two primary approaches: inverse probability weighting (IPW) and full-likelihood (FL) methods. The study reveals that while IPW faces limitations with small target samples, FL can achieve significant efficiency gains, even estimating model parameters at a rate comparable to the auxiliary sample size. The theoretical underpinnings of this 'full efficiency gain' are examined for exponential families and their mixtures, with a discussion on applying FL to neural network models for prognosis. AI

IMPACT Introduces novel statistical techniques that could improve the efficiency of machine learning models trained on limited data.

RANK_REASON The item is an academic paper detailing statistical methodology. [lever_c_demoted from research: ic=1 ai=1.0]

Read on arXiv stat.ML →

• arXiv
• Hugging Face
• inverse probability weighting
• mixtures of exponential families
• Neural network models for breast cancer prognosis
• Tukey

• paper
• other

AI-generated summary · Google Gemini · from 1 sources. How we write summaries →