Plug-and-play (PnP)

Plug-and-play (PnP) is a recent paradigm for signal reconstruction from incomplete noisy measurements that has shown lot of promise. PnP consists of repeatedly inverting a forward model followed by denoising; a powerful denoiser is used which acts as a regularizer. In recent work [1], [2], we prove that for a class of linear denoisers it is possible to come up with  theoretical guarantees on exact and stable recovery of the ground-truth signal. In particular, for compressed sensing using random  matrices, we are able to work out the sample complexity for exact recovery.

References:

[1] R. G. Gavaskar and K. N. Chaudhury, “Regularization using denoising: Exact and robust signal recovery,” Proc. IEEE International Conference on Acoustics, Speech and Signal Processing, 2022.

[2] R. G. Gavaskar, C. D. Athalye and K. N. Chaudhury, “Exact and robust compressed sensing using plug-and-play regularization,” submitted to IEEE Transactions on Image Processing.