AI & ML
Asymptotic efficiency comparison of explicit Dirichlet shape estimators
arXiv stat.TH0 views
arXiv:2611.26124v1 Announce Type: new Abstract: In this study, we compare the asymptotic efficiency of three closed-form estimators of the Dirichlet shape parameters. We start by proving that, in the beta distribution, the score-adjusted moment estimator (SA) has strictly smaller componentwise asymptotic variances than the usual moment estimator (M1) for both parameters. We then introduce a new moment estimator (M2) that combines the marginal moment equations to obtain a common estimate of the total concentration. M2 coincides with M1 in the beta case and has strictly smaller componentwise asymptotic variances for every positive shape vector in three or more dimensions. We then show that SA is asymptotically more efficient than M2 when all shape parameters are equal; if at least two parameters differ, each estimator has smaller asymptotic variance than the other for some linear combination of the parameters. Furthermore, SA has strictly smaller componentwise asymptotic variances than M2 when all shape parameters are at least 1. Finally, we show that this ordering can reverse if a parameter becomes too small, and obtain relevant mathematical and computational bounds. Overall, the study shows that, for the Dirichlet distribution and under a broad range of parameter settings, score-adjusted estimation outperforms the classic moment estimation and identifies the limits of that advantage.
Source:https://arxiv.org/abs/2611.26124PreviousSharp minimax rates and adaptive estimation of latent proportions in bivariate Poisson mixturesNextA Local False Discovery Rate Framework for Compound p-values
Related posts
Why an Embedded Game Fails on Some Phones, and How to Debug It
0dev.to
Microsoft 365 vs Office 2024: The Break-Even Math
0dev.to
Master AI Evals: Build Reliable, Scalable Model Evaluation Pipelines
0dev.to
What to give someone who misses a place
0dev.to
Google is pushing OEMs against GrapheneOS and bootloader unlocking
0Hacker News
VESTA: Visual Exploration with Statistical Tool Agents
0arXiv stat.CO