Kinetic Sunyaev Zel’dovich Velocity Reconstruction Using Vision Transformers

Kinetic Sunyaev Zel'dovich (kSZ) velocity reconstruction uses correlations between maps of the cosmic microwave background (CMB) and a tracer of large scale structure (LSS) to infer the large-scale peculiar velocity field. Standard approaches to kSZ velocity reconstruction use quadratic estimators that are Fisher optimal for purely Gaussian fields. Here, we develop a pipeline for kSZ velocity reconstruction based on Vision Transformers (ViT) and show that this approach can utilize non-Gaussian information in the inputs to improve the fidelity of velocity reconstruction.