AI enhanced solutions for cosmology and astronomy. This work is supported by NSF-Simons AI Institute for Cosmic Origins. For more details please visit cosmicai.org and NSF News article.
Following are the current ongoing research projects:
A novel Cloud-based Astronomy Inference (CAI) framework for data parallel AI model inference on AWS. We can classify 500K astronomy images using the AstroMAE model in a minute !
The first foundation model to process variable lengths of spectra from arbitrary instruments at their original size/resolution, without resizing or downsampling to a fixed size.
Pixel-based diffusion model for reconstructing cosmological initial conditions from Dark Matter Halos. This currently uses the main architecture from the Posterior Sampling of the Initial Conditions of the Universe from Non-Linear Large Scale Structures Using Score-Based Generative Models paper and optimizes it for a new observation dataset. It is trained on Quijote Latin Hypercube simulation dataset to reconstruct the initial condition (z = 127) from DM halo (z=0).
