Python package for scalable Bayesian uncertainty quantification with data-driven priors for radio interferometric imaging in PyTorch
quantifai is a PyTorch-based open source radio interferometric imaging reconstruction package with scalable Bayesian uncertainty quantification relying on data-driven (learned) priors. This was used to produce the results of the paper (Liaudat et al. 2023). The quantifai model relies on the data-driven convex regulariser from the paper (Goujon et al. 2022).
In this code, we bypass the need of to perform Markov chain Monte Carlo (MCMC) sampling for Bayesian uncertainty quantification and we rely on convex accelerated optimisation algorithms. The quantifai package also includes MCMC algorithms for posterior sampling as they were used to validate our approach.
Basic usage is highlighted in this interactive demo.
Please cite Liaudat et al. (2023) if this code package has been of use in your project.
A BibTeX entry for the paper is:
@article{quantifai,
author = {Tobías~I.~Liaudat and Matthijs~Mars and Matthew~A.~Price and Marcelo~Pereyra and Marta~M.~Betcke and Jason~D.~McEwen},
title = {Scalable Bayesian uncertainty quantification with data-driven priors for radio interferometric imaging},
journal = {RAS Techniques and Instruments (RASTI), submitted},
eprint = {arXiv:0000.00000},
year = 2023
}
quantifai is released under the GPL-3 license (see LICENSE.txt).