Neural Parameter Estimation with Incomplete Data

Matthew Sainsbury-Dale, Andrew Zammit-Mangion, Noel Cressie, Raphaël Huser · 2025

Read Neural Parameter Estimation with Incomplete Data by Matthew Sainsbury-Dale, Andrew Zammit-Mangion, Noel Cressie, Raphaël Huser (2025) 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-2501-04330.txt.

Description

Matthew Sainsbury-Dale and three colleagues present a 2025 statistics paper teaching neural networks to do inference when some data are missing. Neural estimators promise instant answers for complex models, but holes in the data break the usual recipe. The easy patch, feeding the network a mask of the holes, gives biased, wasteful results that depend on why data went missing.

Their alternative wraps the network in an expectation-maximization loop done by Monte Carlo, skipping the slow optimization the classical algorithm needs at every step. On simulated spatial data the method beat masking in efficiency, and it handled a real Arctic sea-ice analysis through a hidden Potts model whose likelihood cannot even be written down. The authors pitch it as Bayesian thinking improving AI.

Catalog record

CREATORMatthew Sainsbury-Dale, Andrew Zammit-Mangion, Noel Cressie, Raphaël Huser
YEAR2025
COLLECTIONPAPERS
FORMATTXT
FILE SIZE307 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2501-04330

Browse the full library · About the Tower of Babel