SBI for Denoising?

[Millet21]

Hey everyone,
I was wondering, could SBI also be applied to a data denoising procedure? Indeed, let's assume we have a noisy dataset, a tunable denoising algorithm with some parameters, as well as a "modelled" response that we want to extract from the noisy data. We could then run SBI : we sample parameters for the denoising algorithm and train an NDE to match the denoised data with the synthetic ideal response, thus fine-tuning the denoising procedure. This would be a way to reverse the original paradigm by trying to "fit data to a model" :)
Thanks in advance for any feedback!

[michaeldeistler]

Hi there,

yes, I think that's possible. I have never tried anything like this, and I am not aware of others having worked on something like this, but I think it can definitely work.

Michael