Reducing cross-sample prediction churn in scientific machine learning

TL;DR

This paper investigates the issue of cross-sample prediction churn in scientific machine learning, showing that data-side methods like bootstrap bagging and twin-bootstrap significantly reduce churn compared to standard parameter-side techniques.

Contribution

It introduces twin-bootstrap, a novel training method that reduces prediction churn, and demonstrates the effectiveness of data-side methods over standard techniques.

Findings

Bootstrap bagging reduces churn by 40-54% without accuracy loss.

Twin-bootstrap further reduces churn by median 45% beyond bagging-2.

Standard parameter-side methods do not reduce cross-sample prediction churn.

Abstract

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.