Evaluating transfer across modalities in autonomous materials systems

Read full text Summarise with VIRAT
Authors: Kinston Ackölf, Taylor D. Sparks
Cited by: 0
References: 10

Summary

No summary yet. VIRAT can read this paper and write a plain-language summary of its aim, methods, key findings and limitations.

Summarise with VIRAT Free to try with your Science Catalogs account.

Abstract

Autonomous materials workflows increasingly move between physical models, calibrated simulations, historical datasets, and machine-learning surrogates before decisions are validated in experiment or manufacturing. A central challenge is determining what knowledge remains reusable when the workflow representation changes. Here we introduce PyroEnGarde-Kintsugi (PEGKi), a framework for evaluating transfer across knowledge sources in autonomous materials systems. Using a controlled benchmark in which the nominal task is fixed while the underlying representation changes, we show that transfer is not a simple yes-or-no property. Instead, some forms of knowledge remain portable while others do not, and source usefulness depends on how a representation structures the accessible decision landscape. The results support a more philosophical view of autonomous materials design in which prior knowledge should be treated as partial, context-dependent, and diagnostically valuable rather than universally reusable. PEGKi is therefore best understood as a framework for source selection, transfer diagnosis, and trust in autonomous materials workflows.

Fields of Study & Subjects

Machine Learning in Materials Science Scientific Computing and Data Management Material Selection and Properties

Publication details

Journal / source
ChemRxiv
Publication date
26 March 2026
Language
EN
Document type
Preprint