Birds of a Feather Archive

Building the Scientific Data Fabric Community for AI-Driven Discovery


Authors: Michela Taufer (University of Tennessee, Knoxville; Global Computing Laboratory), Valerio Pascucci (University of Utah), Carol Song (Purdue University), Virginia Trueheart (Texas Advanced Computing Center (TACC)), David Kanter (MLCommons)

Abstract: Artificial intelligence is transforming scientific discovery, but today's cyberinfrastructure was designed for workflows in which data are collected, stored, analyzed, and archived as largely independent activities. Modern science increasingly requires continuous interaction among instruments, data repositories, workflows, AI services, edge devices, HPC systems, and scientists across institutional boundaries. This BoF brings together researchers, facility operators, cyberinfrastructure developers, and domain scientists to discuss the next generation of scientific data infrastructure. Through examples from astronomy, geosciences, materials science, medicine, and high-energy physics, participants will identify common infrastructure challenges and define a community roadmap for interoperable, federated scientific data ecosystems.

Long Description: Artificial intelligence is changing how science is conducted. Across disciplines, AI is increasingly integrated into observation, simulation, analysis, and decision making. Scientific facilities are becoming more connected, experiments generate continuous streams of data, and collaborations increasingly span institutions, facilities, cloud platforms, and HPC centers. While computing capabilities have advanced dramatically, the underlying scientific data infrastructure has evolved much more slowly.

Today's cyberinfrastructure remains fragmented. Data repositories, workflow systems, storage platforms, scientific instruments, metadata catalogs, visualization services, and AI frameworks are often developed independently, making it difficult to build end-to-end scientific workflows that operate across organizational boundaries. These limitations affect a broad range of communities, including astronomy, high-energy physics, geosciences, materials science, neutron and light sources, medicine, climate science, and national research facilities.

This BoF asks a fundamental question:

What should scientific data infrastructure look like in the era of AI-driven discovery?

Rather than focusing on any particular architecture or software platform, the discussion will identify the common capabilities required by future scientific cyberinfrastructure. Topics include federated data access, distributed metadata management, provenance, persistent digital objects, policy-aware data sharing, workflow integration, AI-ready data services, edge-to-cloud computing, interoperability across facilities, and long-term stewardship of scientific data.

The session will begin with several short case studies illustrating common challenges across different scientific domains, including autonomous beamline experiments, distributed observational science, large-scale simulation campaigns, medical data ecosystems, and data-intensive astronomy. Although these communities have different scientific objectives, they increasingly require the same foundational cyberinfrastructure to support collaborative and AI-enabled discovery.

The majority of the session will be devoted to an interactive discussion in which participants identify shared requirements, existing gaps, emerging standards, and opportunities for collaboration across communities. Rather than developing another domain-specific solution, the goal is to articulate a common vision for scientific data infrastructure that enables interoperable, trustworthy, and scalable scientific discovery.

The outcome of the BoF will be a community-driven roadmap identifying research priorities, technology gaps, and opportunities for collaboration among the HPC, AI, cyberinfrastructure, and scientific communities.

Website: https://nationalsciencedatafabric.org/



Back to Birds of a Feather Archive Listing