Realizable Bayes-Consistency for General Metric Losses

Fuente: arXiv
Gespeichert in:
Bibliographische Detailangaben
Hauptverfasser: Cohen, Dan Tsir, Hanneke, Steve, Kontorovich, Aryeh
Format: Preprint
Veröffentlicht: 2026
Schlagworte:
Online-Zugang:
Tags: Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
_version_ 1866911685338464256
author Cohen, Dan Tsir
Hanneke, Steve
Kontorovich, Aryeh
author_facet Cohen, Dan Tsir
Hanneke, Steve
Kontorovich, Aryeh
contents We study strong universal Bayes-consistency in the realizable setting for learning with general metric losses, extending classical characterizations beyond $0$-$1$ classification (Bousquet et al., 2020; Hanneke et al., 2021) and real-valued regression (Attias et al., 2024). Given an instance space $(X,ρ)$, a label space $(Y,\ell)$ with possibly unbounded loss, and a hypothesis class $H \subseteq Y^{X}$, we resolve the realizable case of an open problem presented in Tsir Cohen and Kontorovich (2022). Specifically, we find the necessary and sufficient conditions on the hypothesis class $H$ under which there exists a distribution-free learning rule whose risk converges almost surely to the best-in-class risk (which is zero) for every realizable data-generating distribution. Our main contribution is this sharp characterization in terms of a combinatorial obstruction: Similarly to Attias et al. (2024), we introduce the notion of an infinite non-decreasing $(γ_k)$-Littlestone tree, where $γ_k \to \infty$. This extends the Littlestone tree structure used in Bousquet et al. (2020) to the metric loss setting.
format Preprint
id arxiv_https___arxiv_org_abs_2605_03823
institution arXiv
publishDate 2026
record_format arxiv
spellingShingle Realizable Bayes-Consistency for General Metric Losses
Cohen, Dan Tsir
Hanneke, Steve
Kontorovich, Aryeh
Machine Learning
Information Theory
Statistics Theory
68Q32, 68T05, 62G08, 62G20
I.2.6; F.2.0; G.3
We study strong universal Bayes-consistency in the realizable setting for learning with general metric losses, extending classical characterizations beyond $0$-$1$ classification (Bousquet et al., 2020; Hanneke et al., 2021) and real-valued regression (Attias et al., 2024). Given an instance space $(X,ρ)$, a label space $(Y,\ell)$ with possibly unbounded loss, and a hypothesis class $H \subseteq Y^{X}$, we resolve the realizable case of an open problem presented in Tsir Cohen and Kontorovich (2022). Specifically, we find the necessary and sufficient conditions on the hypothesis class $H$ under which there exists a distribution-free learning rule whose risk converges almost surely to the best-in-class risk (which is zero) for every realizable data-generating distribution. Our main contribution is this sharp characterization in terms of a combinatorial obstruction: Similarly to Attias et al. (2024), we introduce the notion of an infinite non-decreasing $(γ_k)$-Littlestone tree, where $γ_k \to \infty$. This extends the Littlestone tree structure used in Bousquet et al. (2020) to the metric loss setting.
title Realizable Bayes-Consistency for General Metric Losses
topic Machine Learning
Information Theory
Statistics Theory
68Q32, 68T05, 62G08, 62G20
I.2.6; F.2.0; G.3
url https://arxiv.org/abs/2605.03823