Team

Large astronomical datasets, which are by nature public and non-proprietary, provide unique opportunities to advance AI research in a “safe” public domain. Our Explorable Universe group aims to advance generative AI trustworthiness and reasoning by developing open multi-modal large language models (LLMs) for Astronomy.

Astronomical data, which are inherently complex, noisy, and high-dimensional, provide challenging use cases to improve the automation of data preparation. The Observable Universe group aims to develop efficient techniques for the calibration and analysis of high-dimensional data, and to automate data pre-processing of interferometric spectral image cubes, which will increase by two orders of magnitude in size within a decade.

AI-enabled regression and inference techniques have unlocked massive discovery potential; however, constructing accurate predictions is inadequate without an understanding of the inference process itself. The Explainable Universe group aims to extend interpretable AI methods and unify them with causal reasoning, enabling explanation of the underlying mathematical relationships, while remaining robust to noise and uncertainty.

AI advances in simulating the behavior of physical systems provide a means to significantly accelerate the speed of astrophysical calculations and gain new insights. The Accelerated Universe group aims to extend the robustness of AI surrogates in order to model multi-physics, multiscale problems with the large dynamic ranges characteristic of astrophysical systems.