Generalized resilience and robust statistics
نویسندگان
چکیده
Robust statistics traditionally focuses on outliers, or perturbations in total variation distance. However, a dataset could be maliciously corrupted many other ways, such as systematic measurement errors and missing covariates. We consider corruption either TV Wasserstein distance, show that robust estimation is possible whenever the true population distribution satisfies property called generalized resilience, which holds under moment hypercontractive conditions. For model, our finite-sample analysis improves over previous results for mean with bounded kth moment, linear regression, joint covariance estimation. W1 corruption, we provide first guarantees second regression. Technically, estimators are generalization of minimum distance (MD) functionals, project onto given set well-behaved distributions. The error these MD functionals by certain modulus continuity, method upper bounding this class resilient distributions, usually gives sharp population-level good guarantees.
منابع مشابه
Statistics of Robust Optimization: A Generalized Empirical Likelihood Approach
We study statistical inference and robust solution methods for stochastic optimization prob-lems. We first develop an empirical likelihood framework for stochastic optimization. We showan empirical likelihood theory for Hadamard differentiable functionals with general f -divergencesand give conditions under which T (P ) = infx∈X EP [`(x; ξ)] is Hadamard differentiable. Noting<lb...
متن کاملMultivariate Dispersive Ordering of Generalized Order Statistics
The concept of generalized order statistics (GOSs) was introduced as a unified approach to a variety of models of ordered random variables. The purpose of this paper is to investigate condi- tions on the underlying distribution functions and the parameters on which GOSs are based, to establish Shaked-Shanthikumar multivari- ate dispersive ordering of GOSs from one sample and Khaledi-Kochar mu...
متن کاملRecurrence Relations for Quotient Moment of Generalized Pareto Distribution Based on Generalized Order Statistics and Characterization
Generalized Pareto distribution play an important role in reliability, extreme value theory, and other branches of applied probability and statistics. This family of distributions includes exponential distribution, Pareto distribution, and Power distribution. In this paper, we established exact expressions and recurrence relations satisfied by the quotient moments of generalized order statistic...
متن کاملAmbiguity and robust statistics
Since the seminal work of Gilboa and Schmeidler [28, p. 142] a relation between decision making under ambiguity and robust Bayesian statistics has been hinted at, and indeed immediate similarities are quite evident. At the same time, a formal treatment of this topic and a complete characterization of the relation between the two approaches is still missing. The object of this paper is to ll th...
متن کاملRobust statistics
Robust stat ist ics seeks to provide methods that emulate popular stat ist ical methods, but which are not unduly af fected by out liers or other small departures f rom model assumptions. In stat ist ics, classical est imat ion methods rely heavily on assumptions which are of ten not met in pract ice. In part icular, it is of ten assumed that the data errors are normally distributed, at least a...
متن کاملذخیره در منابع من
با ذخیره ی این منبع در منابع من، دسترسی به آن را برای استفاده های بعدی آسان تر کنید
ژورنال
عنوان ژورنال: Annals of Statistics
سال: 2022
ISSN: ['0090-5364', '2168-8966']
DOI: https://doi.org/10.1214/22-aos2186