AI/ML STIG

NASA is establishing a new AI/ML Science and Technology Interest Group (STIG) under the Cosmic Origins Program Analysis Group (COPAG). COPAG is an open community forum supporting NASA's Cosmic Origins Program. STIGs are long-term affinity groups that advance specific subfields through regular meetings and knowledge sharing.

This initiative comes at a critical time. The recent community paper on the Future of AI and Mathematical and Physical Sciences (https://arxiv.org/abs/2509.02661) identified the need to upskill the scientific community with better AI literacy, enabling us to make informed decisions about—and constructive critiques of—AI's role in astronomy. To maximize the science returns from NASA's missions and accelerate scientific discoveries, we need stackable modular training specifically designed for astronomy. While astronomy researchers are becoming familiar with AI as statistical tools, it remains extremely difficult to find systematic resources about AI relevant to astronomical research contexts or formats that accommodate our time constraints.

For our first year, the AI/ML STIG aims to address this gap by delivering a ~26-week series of 40-minute sessions—community-curated content addressing specific astronomical AI applications. Our tentative activities (November 1, 2025 - May 31, 2026, with winter break) include:

• Weeks 1-3: Large Language Models as Research Agents [Tutorials]
• Week 4: How to get funding/compute from NASA/NSF Resources for AI research [Seminar]
• Weeks 5-6: Modern GPU-Friendly Frameworks - PyTorch, JAX [Tutorials]
• Week 7: Neural Network Theory - Inductive Biases [Seminars]
• Weeks 8-11: Neural Network Architectures - CNNs, RNNs, GNNs, Transformers [Tutorials]
• Weeks 12-13: Equivariant Network Theory and Applications [Tutorial]
• Week 14: Mid-Series Town Hall - where AI might play the biggest role [Townhall]
• Weeks 15-19: Generative Models as Neural Density Estimators - Normalizing Flows, Diffusion Models, Flow Matching [Tutorials/Seminars]
• Week 20: Simulation-Based Inference and Field Level Inference [Seminar]
• Weeks 21-23: Reinforcement Learning Implementation and Application [Tutorials/Seminars]
• Week 24: Open Benchmarking Datasets [Seminar]
• Week 25: Foundation Models [Seminar]
• Week 26: Final Town Hall - Bring Your Questions [Townhall]

Action Required

1. Sign up for the mailing list

2. Nominate speakers or self-nominate for topics by Oct 10 and suggest additional AI/ML topics you'd like to see covered. Your suggestions will help shape future sessions and, where possible, may be incorporated into this year's program: NASA AI-ML STIG Speakers Nomination Form

We will send a poll to determine optimal meeting times for those on the mailing list.

cheers,

Yuan-Sen Ting (OSU)

On behalf of the AI/ML STIG Leadership Council:

Alex Gagliano (MIT)

Siddharth Mishra-Sharma (Boston University)

Digvijay Wadekar (Johns Hopkins)

Andrew Saydjari (Princeton)

Carol Cuesta-Lazaro (MIT)

Georgios Valogiannis (UChicago)