Convergence rates for the generalized Fréchet mean via the quadruple inequality
For sets QQ\mathcal{Q} and YY\mathcal{Y}, the generalized Fréchet mean m∈Qm∈Qm\in \mathcal{Q} of a random variable YYY, which has values in YY\mathcal{Y}, is any minimizer of q↦E[c(q,Y)]q↦E[c(q,Y)]q\mapsto \mathbb{E}[\mathfrak{c}(q,Y)], where c:Q×Y→Rc:Q×Y→R\mathfrak{c}\colon \mathcal{Q}\times \math...
Saved in:
| Main Author: | |
|---|---|
| Format: | Article (Journal) |
| Language: | English |
| Published: |
2019
|
| In: |
Electronic journal of statistics
Year: 2019, Volume: 13, Issue: 2, Pages: 4280-4345 |
| ISSN: | 1935-7524 |
| DOI: | 10.1214/19-EJS1618 |
| Online Access: | Verlag, Volltext: https://doi.org/10.1214/19-EJS1618 Verlag, Volltext: https://projecteuclid.org/euclid.ejs/1572249628 
|
| Author Notes: | Christof Schötz |
| Summary: | For sets QQ\mathcal{Q} and YY\mathcal{Y}, the generalized Fréchet mean m∈Qm∈Qm\in \mathcal{Q} of a random variable YYY, which has values in YY\mathcal{Y}, is any minimizer of q↦E[c(q,Y)]q↦E[c(q,Y)]q\mapsto \mathbb{E}[\mathfrak{c}(q,Y)], where c:Q×Y→Rc:Q×Y→R\mathfrak{c}\colon \mathcal{Q}\times \mathcal{Y}\to \mathbb{R} is a cost function. There are little restrictions to QQ\mathcal{Q} and YY\mathcal{Y}. In particular, QQ\mathcal{Q} can be a non-Euclidean metric space. We provide convergence rates for the empirical generalized Fréchet mean. Conditions for rates in probability and rates in expectation are given. In contrast to previous results on Fréchet means, we do not require a finite diameter of the QQ\mathcal{Q} or YY\mathcal{Y}. Instead, we assume an inequality, which we call quadruple inequality. It generalizes an otherwise common Lipschitz condition on the cost function. This quadruple inequality is known to hold in Hadamard spaces. We show that it also holds in a suitable way for certain powers of a Hadamard-metric. |
|---|---|
| Item Description: | Gesehen am 27.01.2020 |
| Physical Description: | Online Resource |
| ISSN: | 1935-7524 |
| DOI: | 10.1214/19-EJS1618 |