Matthew Craigie, Eric Huff, Yuan-Sen Ting, Rossana Ruggeri, Tamara M. Davis

We present DELTA (Data-Empiric Learned Tidal Alignments), a deep learning model that isolates galaxy intrinsic alignments (IAs) from weak lensing distortions using only observational data. The model uses an Equivariant Graph Neural Network backbone suitable for capturing information from the local galaxy environment, in conjunction with a probabilistic orientation output. Unlike parametric models, DELTA flexibly learns the relationship between galaxy shapes and their local environments, without assuming an explicit IA form or relying on simulations. When applied to mock catalogs with realistic noisy IAs injected, it accurately reconstructs the noise-free, pure IA signal. Mapping these alignments provides a direct visualization of IA patterns in the mock catalogs. Combining DELTA with deep learning interpretation techniques provides further insights into the physics driving tidal relationships between galaxies. This new approach to understanding and controlling IAs is suitable for application to joint photometric and spectroscopic surveys such as the combination of upcoming Euclid, Rubin, and DESI datasets.