Reducing cross-sample prediction churn in scientific machine learning

Fuente: arXiv
Saved in:
Bibliographic Details
Main Authors: Prastalo, Gordan, Jablonka, Kevin Maik
Format: Preprint
Published: 2026
Subjects:
Online Access:
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866913124429332480
author Prastalo, Gordan
Jablonka, Kevin Maik
author_facet Prastalo, Gordan
Jablonka, Kevin Maik
contents Scientific machine learning reports predictive performance. It does not report whether the same prediction would survive a different draw of training data. Across $9$ chemistry benchmarks, two classifiers trained on independent bootstraps of the same training set agree on aggregate accuracy to within $1.3\text{--}4.2$ percentage points but disagree on the class label of $8.0\text{--}21.8\%$ of test molecules. We call this gap \emph{cross-sample prediction churn}. The standard parameter-side techniques (deep ensembles, MC dropout, stochastic weight averaging) do not reduce this gap; two data-side methods do. The first is $K$-bootstrap bagging, which cuts the rate $40\text{--}54\%$ on every dataset at no accuracy cost ($K{\times}$-ERM compute). The second is \emph{twin-bootstrap}, our proposal: two networks trained jointly on independent bootstraps with a sym-KL consistency loss between their predictions, which at matched $2{\times}$-ERM compute reduces churn a further median $45\%$ beyond bagging-$K{=}2$. Cross-sample prediction churn deserves a column alongside predictive performance in scientific-ML benchmark reports, because without it the parameter-side and data-side methods are indistinguishable on the metric they actually differ on.
format Preprint
id arxiv_https___arxiv_org_abs_2605_13826
institution arXiv
publishDate 2026
record_format arxiv
spellingShingle Reducing cross-sample prediction churn in scientific machine learning
Prastalo, Gordan
Jablonka, Kevin Maik
Machine Learning
Materials Science
Chemical Physics
Scientific machine learning reports predictive performance. It does not report whether the same prediction would survive a different draw of training data. Across $9$ chemistry benchmarks, two classifiers trained on independent bootstraps of the same training set agree on aggregate accuracy to within $1.3\text{--}4.2$ percentage points but disagree on the class label of $8.0\text{--}21.8\%$ of test molecules. We call this gap \emph{cross-sample prediction churn}. The standard parameter-side techniques (deep ensembles, MC dropout, stochastic weight averaging) do not reduce this gap; two data-side methods do. The first is $K$-bootstrap bagging, which cuts the rate $40\text{--}54\%$ on every dataset at no accuracy cost ($K{\times}$-ERM compute). The second is \emph{twin-bootstrap}, our proposal: two networks trained jointly on independent bootstraps with a sym-KL consistency loss between their predictions, which at matched $2{\times}$-ERM compute reduces churn a further median $45\%$ beyond bagging-$K{=}2$. Cross-sample prediction churn deserves a column alongside predictive performance in scientific-ML benchmark reports, because without it the parameter-side and data-side methods are indistinguishable on the metric they actually differ on.
title Reducing cross-sample prediction churn in scientific machine learning
topic Machine Learning
Materials Science
Chemical Physics
url https://arxiv.org/abs/2605.13826