PHYSICS-INFORMED MACHINE LEARNING FOR WEAK GRAVITATIONAL LENSING COSMOLOGY

Our universe is filled with mysteries we cannot see directly, such as dark matter and dark energy. Together, these make up about 95% of the cosmos, yet we only know them through their indirect effects. One of the most powerful tools to study them is called weak gravitational lensing. When light from very distant galaxies travels through the universe, it is very slightly bent by the gravity of matter along its path. This causes tiny distortions in the apparent shapes of those galaxies. By measuring these subtle shape changes across millions of galaxies, astronomers can create maps of the hidden dark matter and better understand how the universe is expanding.

Detecting such faint signals is extremely challenging. The distortions are so small that even tiny errors in how we measure galaxy shapes can mislead us. Professor Liu’s project will tackle this problem by developing a new kind of computer algorithm based on artificial intelligence (AI). Unlike traditional AI approaches that act like “black boxes,” this model will be physics-informed—built with the known symmetries of the universe in mind. For example, rotating or mirroring a galaxy image does not change the physics, and the algorithm should recognize that. By embedding these symmetries directly into the design, the method becomes more accurate and trustworthy. Moreover, the algorithm will be able to self-check its own bias using modern techniques from machine learning.

This project is especially timely because the Vera C. Rubin Observatory’s Legacy Survey of Space and Time (LSST)—the highest-priority ground-based astronomy project—is about to begin operations. LSST will capture billions of galaxies in unprecedented detail, producing an enormous dataset. By preparing new AI tools now, researchers will be ready to fully harness this upcoming flood of data. This fellowship semester will provide the dedicated time to build and test a first prototype, laying the foundation for future discoveries about the unseen universe.