Discovering new patterns in cosmological data with interpretable deep learning
I demonstrate how the direct application of deep learning to cosmology datasets can provide a pathway to discovering new, previously unrecognized patterns. I present two examples of this approach: the search for mirror-asymmetric structures in the galaxy distribution, and the modeling of galaxy alignments due to tidal interactions. In these examples, I emphasize the importance of using physically-motivated deep learning models and emerging interpretability techniques to ensure transparent and trustworthy science. This work shows how deep learning can be used not just for modeling known relationships in data, but also for identifying new ones.
For Zoom information, please contact the seminar coordinators.