New AI Models Unlock Hidden Patterns in Stars and Fluid Dynamics

Researchers at the Flatiron Institute and Polymathic AI have developed two new physics-driven foundation models. These tools, named AION-1 and Walrus, learn the underlying mathematical patterns in complex physical systems. Their ability to reconstruct missing data could transform how scientists analyze everything from stars to fluid dynamics.

The first model, AION-1, specializes in astronomy. It processes vast collections of star images and spectral data to uncover hidden relationships. By identifying these patterns, it allows astronomers to apply insights from one area of study to entirely different cosmic phenomena.

The second model, Walrus, focuses on fluid and fluid-like systems. It examines behavior in small-scale experiments and then predicts how similar forces play out in extreme environments. For example, it can infer the dynamics of stellar explosions by studying lab-based fluid flows.

Both models share a core strength: they reconstruct incomplete or obscured scientific measurements. Instead of relying on pre-programmed equations, they detect the mathematical structure of physical systems on their own. This approach lets them bridge gaps between different scales and conditions, offering researchers a more flexible way to explore uncharted territory.

These models mark a step toward broader applications of AI in physics. AION-1's adaptability in astronomy and Walrus's ability to scale fluid dynamics insights could help scientists tackle problems that were previously too complex or data-scarce. While institutions like ILM Ulm and Fraunhofer SCAI also work on related computational research, the scope of these new tools extends beyond single disciplines.