SMRS: advocating a unified reporting standard for surrogate models in the artificial intelligence era.

Elizaveta Semenova, Siobhan Mackenzie Hall, Timothy James Hitge, Alisa Sheinkman, Jon Cockayne

Keywords: surrogates, evaluation, sampling design, model selection, reporting standard

TL;DR: This paper advocates for a unified framework for surrogate model training and evaluation to address methodological fragmentation, improve reproducibility, and enable cross-domain collaboration in scientific research.

Abstract:

Surrogate models are widely used to approximate complex systems across science and engineering to reduce computational costs. Despite their widespread adoption, the field lacks standardisation across key stages of the modelling pipeline, including data sampling, model selection, evaluation, and downstream analysis. This fragmentation limits reproducibility and cross-domain utility – a challenge
further exacerbated by the rapid proliferation of AI-driven surrogate models. We argue for the urgent need to establish a structured reporting standard, the Surrogate Model Reporting Standard (SMRS), that systematically captures essential design and evaluation choices while remaining agnostic to implementation specifics. By promoting a standardised yet flexible framework, we aim to improve the reliability
of surrogate modelling, foster interdisciplinary knowledge transfer, and, as a result, accelerate scientific progress in the AI era.

Lay Summary:

Scientific and engineering problems, from climate prediction to drug discovery, often depend on surrogate models: simplified, data-driven models that approximate expensive simulations or experiments. These surrogates save enormous amounts of time and computation but are developed in very different ways across disciplines. As a result, it is often impossible to reproduce, compare, or reuse them, even when code and data are available.

This paper argues that the field needs a shared way to report how surrogate models are built, trained, and tested. We introduce the Surrogate Model Reporting Specification (SMRS), a lightweight checklist that captures key design choices, such as how data were collected, what type of model was used, how uncertainty was handled, and how performance was evaluated. SMRS is not a new algorithm; it is a structured way to describe existing ones so that other researchers can understand and replicate them.

The paper outlines the main elements of the specification and demonstrates how it applies to recent studies in areas such as climate science, geophysics, and neuroscience. We also discuss concerns about over-standardisation and explain how SMRS remains flexible and domain-adaptable.

By introducing a consistent reporting standard, we aim to make surrogate modelling more transparent, comparable, and trustworthy, ultimately accelerating progress across many scientific fields in the age of artificial intelligence.

Submission Number: 97