Open dfm opened 2 years ago
Exciting to watch the development of this repo and know that you are thinking about machine learning applications :) 2010.05295 is a very relevant reference for that. Interested to know what applications do you have in mind.
Thanks @eelregit! I must admit that none of my use cases would really be counted as "Machine Learning". In fact, I don't have immediate applications, but I'd be excited to use this for probabilistic models of interferometric data (@iancze has been thinking about such things), optical design (cc @benjaminpope), and I know that @davidwhogg has some spectroscopy related applications, but I don't know all the details.
For this issue, we're considering the possibility that there might be scope for a publication without tying to a specific application, since we've done/are doing some useful code (for example, support for CUDA streams) and method development (defining the differentiation rules).
Thanks @dfm. Looking forward to using the awesome jax interface!
PS for non-ML applications, our (soon to be open) PM code can also benefit from this project for more accurate long-range forces.
Just to tag along, I would love to use this as part of a differentiable plasma physics simulation once the GPU bindings are implemented. There is definitely some interesting physics to explore!
I should also be more useful though, here are some journal suggestions - How about
@eelregit @joglekara Just to circle back on this (after quite a while!), the GPU JAX bindings are now available if either of you would like to try them out.
@lgarrison and I have been chatting about the possibility of writing a paper describing what we're doing here. Something like "Differentiable programming with NUFFTs" or "NUFFTs for machine learning applications" or something. This issue is here to remind me to look into possible venues for this.