MIRA: A Score for Conditional Distribution Accuracy and Model Comparison

Sammy Sharief, Justine Zeghal, Gabriel Missael Barco, Pablo Lemos, Yashar Hezaveh, Laurence Perreault-Levasseur · 2026

Read MIRA: A Score for Conditional Distribution Accuracy and Model Comparison by Sammy Sharief, Justine Zeghal, Gabriel Missael Barco, Pablo Lemos, Yashar Hezaveh, Laurence Perreault-Levasseur (2026) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2605-02014.txt.

Description

Sammy Sharief and colleagues introduce Mira, a score for checking how accurate a proposed conditional distribution is. Conditional models predict an outcome given inputs, and judging them normally needs hard evidence calculations. Mira uses only joint samples from the true process and the idea that two distributions match if they give the same mass to every region.

From that idea, the authors derive a closed form for the Mira statistic and its average score. When a candidate matches the truth, they can also compute reference values and uncertainty. Mira thus allows Bayesian model comparison by direct validation of the posterior, bypassing evidence. Tests on toy cases and inference tasks show the score works as intended.

Catalog record

CREATORSammy Sharief, Justine Zeghal, Gabriel Missael Barco, Pablo Lemos, Yashar Hezaveh, Laurence Perreault-Levasseur
YEAR2026
COLLECTIONPAPERS
FORMATTXT
FILE SIZE106 KB
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2605.02014

Browse the full library · About the Tower of Babel