MLExchange - A Collaborative Machine Learning Platform for Scientific Discovery

The Department of Energy’s (DOE) Scientific User Facilities (SUFs) are renowned worldwide for producing high-quality scientific data from various experiments, models, and simulations. The data generated at SUFs is complex and diverse, necessitating specialized expertise for proper interpretation. Scientific machine learning (ML) advancements present a unique opportunity to leverage collective knowledge, shared experiences, and insights from the scientific user facility community. Recent breakthroughs in ML have had a tremendous impact across industry and science domains, primarily due to the readily available diverse data sets on which networks can train to select values for millions of internal parameters required to produce effective models.

MLExchange is a software framework aimed at providing solutions to the challenges encountered by the scientific community in the realm of ML and integrating it into the workflow of SUFs. The framework addresses the challenges associated with effectively applying ML techniques: constructing models, evaluating algorithms, curating data, and disseminating efficient ML frameworks across experiments, facilities, and scientific inquiries. MLExchange has been deployed at instruments across multiple SUFs.

Ongoing work includes enabling more accurate labeling of scientific n-dimensional data, tools to offer researchers greater explainability and interpretability in their work, and prompt, interactive feedback to framework users, facilitating real-time annotation and analysis of extensive data sets. MLExchange will advance the development of materials science-driven machine learning to offer a comprehensive experience encompassing materials development, measurement, and machine learning-driven analysis of these materials across different techniques.