Efficient and scalable implementation of pixelized source reconstructions for interferometer analysis using JAX.
This PR introduces a new approach to source reconstructions for interferometer data that fully exploits the symmetries and sparsity of the non-uniform fast transformation.
A high level summary of the implementation is:
-
Pixelized source reconstructions are performed in a way whereby the run time and amount of VRAM used is independent of the number of visibilities.
-
Lens modeling run times are fast, with a 1+ million visibility ALMA dataset being modeled in around 1 hour on a GPU!
-
Other improvements to interferometer analysis and a significant portion of support documentation and examples on the
autolens_workspaceare now provided.
Whilst a quantitative comparison has not yet been performed, my intuition is that this code runs significantly faster than the previous PyAutoLens interferometer modeling and the Powell et al implementation.
Checkout the interferometer package of the autolens_workspace for a complete run through of how to use JAX GPU interferometer analysis!
https://github.com/Jammy2211/autolens_workspace/tree/release/notebooks/interferometer