leon-bungert / Eigenvectors-of-Proximal-Operators-and-Neural-Networks

Nonlinear Power Method for Computing Eigenvectors of Proximal Operators and Neural Networks

Eigenvectors of Proximal Operators and Neural Networks

This Python and MATLAB code allows to reproduce some of the results of Nonlinear Power Method for Computing Eigenvectors of Proximal Operators and Neural Networks [1]. Feel free to use it and please refer to our paper when doing so.

Prerequistes

The code for the proximal power method is written in Python and requires the Operator Discretization Library (ODL) https://odlgroup.github.io/odl/index.html. The code for the power method for the denoising neural network FFDnet [2] is written in MATLAB.

References

[1] Leon Bungert, Ester Hait-Fraenkel, Nicolas Papadakis, and Guy Gilboa. "Nonlinear Power Method for Computing Eigenvectors of Proximal Operators and Neural Networks." arXiv preprint arXiv:2003.04595 (2020). https://arxiv.org/abs/2003.04595

[2] Kai Zhang, Wangmeng Zuo, and Lei Zhang. "FFDNet: Toward a fast and flexible solution for CNN-based image denoising." IEEE Transactions on Image Processing 27, no. 9 (2018): 4608-4622.