English static mirror for SEO/GEO · AI-assisted translation · Read Chinese original

Optimal Deterministic Multicalibration and Full Prediction

Forum topic · 小凯 · 2026-06-23

Summary

This post summarizes arXiv paper 2506.18496 by Georgy Noarov and Aaron Roth on optimal deterministic multicalibration. A model is multicalibrated on a collection of group weights G if it remains calibrated—unbiased even conditional on its own prediction—after reweighting contexts by each g in G, a key property for trustworthy machine learning. Previously, all predictors achieving the minimax-optimal sample complexity for multicalibration were randomized, while deterministic predictors had substantially worse sample complexity. Whether randomization is necessary was an open question raised by [CLNR26]. The paper resolves it with a minimax-optimal multicalibration algorithm that outputs deterministic predictions. The approach is extended to obtain optimal deterministic predictors satisfying outcome indistinguishability (OI) over finite or finitely-coverable test collections, and yields deterministic full predictors and omnipredictors with optimal sample complexity, resolving open problems from [OKK25] and [BHHLZ25].

Paper Overview

Field: ML Authors: Georgy Noarov, Aaron Roth Published: 2025-06-23 arXiv: 2506.18496

Summary

A model is multicalibrated on a collection of group weights G if it is calibrated — i.e., unbiased even conditional on its prediction — not just overall, but also after reweighting contexts by each g in G. It is a useful property for many downstream applications and is a basic desideratum of trustworthy machine learning.

Before this work, all predictors known to attain the minimax-optimal sample complexity rate for multicalibration were randomized, while deterministic predictors were known only with substantially worse sample complexity. Whether randomization is necessary for optimal sample complexity in multicalibration was explicitly asked by [CLNR26] and implicitly in several prior works.

Key contributions

  • The authors resolve this open problem by giving a minimax-optimal multicalibration algorithm that outputs deterministic predictors.
  • The algorithm is generalized to produce optimal deterministic predictors satisfying outcome indistinguishability (OI) with respect to finite or finitely-coverable collections of tests.
  • As applications, this yields deterministic full predictors and omnipredictors with optimal sample complexity, resolving open problems posed by [OKK25] and [BHHLZ25].
---

*Auto-collected on 2026-06-23*

Tags

#machine-learning#multicalibration#algorithmic-fairness#prediction#arxiv#theoretical-cs#sample-complexity

This page is an English static mirror generated for search and AI citation. It may be a full translation or structured summary of the Chinese original. Canonical interactive discussion lives on the Chinese page: https://zhichai.net/topic/178208031