TODO list

[KhaledNabilSharafeldin]

Create a todo list as a pinned issue.

[brussel13]

Add a Jax implementation to cohlib. Just wanted to close the browser window and not forget about Jax.

[brussel13]

Add algorithms based on automatic differentiation to optimize against a discrepancy model. This could also include noise model. Jax and/or pytorch can do these things.

[brussel13]

Look at using non-uniform FFT. Particularly the multipeak algorithm. Could save the initial rotate/interpolate of the data.
Could also let us phase with the image in a laboratory/uniform frame of reference in the single peak case. So we wouldn't need the after-phasing coordinate transform. Also, it does away with the approximation that we sample uniformly with a flat detector. We would just compute the Q of each pixel in the measurement and give that to the nuFFT.
Could also develop multi-scale algorithms because the sampling in direct (image) space does not need to be uniform. Might be good for a hybrid continuum/atomistic reconstruction algorithm.
Would work for both projection based algorithms (HIO/ER) and automatic differentiation algorithms. Except I don't think Flatiron is differentiable. There might be a version that is though.

Flatiron nuFFT seems really good.

[brussel13]

Add algorithms based on automatic differentiation to optimize against a discrepancy model. This could also include noise model. Jax and/or pytorch can do these things.

Another advantage is no need for the inverse problem. Only need to compute the forward problem from image to reciprocal space. So can include dynamical effects (Longlong paper). Can also use models that emphasize high Q (low intensity) signals.

[brussel13]

Finally got the dislocation mapping algorithm from David Yang. Both a matlab code and a somewhat untested python version developed by a summer student. Also have test data sets. Would be good to try to add this to cohere-ui postprocessing.

[brussel13]

Just capture this. Plan for another implementation of cohlib.
https://ml-explore.github.io/mlx/build/html/index.html