Finite-Sample Maximum Likelihood Estimation of Location
Statistics Theory
2022-07-20 v2 Data Structures and Algorithms
Information Theory
Machine Learning
math.IT
Machine Learning
Statistics Theory
Abstract
We consider 1-dimensional location estimation, where we estimate a parameter from samples , with each drawn i.i.d. from a known distribution . For fixed the maximum-likelihood estimate (MLE) is well-known to be optimal in the limit as : it is asymptotically normal with variance matching the Cram\'er-Rao lower bound of , where is the Fisher information of . However, this bound does not hold for finite , or when varies with . We show for arbitrary and that one can recover a similar theory based on the Fisher information of a smoothed version of , where the smoothing radius decays with .
Cite
@article{arxiv.2206.02348,
title = {Finite-Sample Maximum Likelihood Estimation of Location},
author = {Shivam Gupta and Jasper C. H. Lee and Eric Price and Paul Valiant},
journal= {arXiv preprint arXiv:2206.02348},
year = {2022}
}
Comments
Corrected an inaccuracy in the description of the experimental setup. Also updated funding acknowledgements