automatic differentiation support?

roflmaostc / DeconvOptim.jl

A multi-dimensional, high performance deconvolution framework written in Julia Lang for CPUs and GPUs.

https://roflmaostc.github.io/DeconvOptim.jl/stable/

MIT License

60 stars 6 forks source link

automatic differentiation support? #51

Open stevengj opened 3 months ago

stevengj commented 3 months ago

It might be nice to add support for ImplicitDifferentiation.jl or similar to make the reconstructed image differentiable with respect to the input image (and/or the hyper-parameters like λ).

@gaurav-arya worked out the algorithms for this in this paper (2022), section 2.2 (for ElasticNet regularization, which includes TV and Tikhonov as special cases), where he used it to co-design optics for CS reconstruction.

roflmaostc commented 3 months ago

I would need to read on that.

Can @gaurav-arya help regarding the implementation?