Plug-and-Play Unplugged: Optimization Free Regularization using Consensus EquilibriumMS55

Recently, the Plug-&-Play method was introduced as a way to use advanced non-parametric denoising algorithms such as BM3D and CNNs as prior models in inverse problems. In this paper, we introduce a generalization of P&P which we call Multi-Agent Consensus Equilibrium (MACE). MACE addresses two limitations of P&P. First, it allows for the introduction of multiple prior or data terms or agents. These agents may, for example, represent multiple uncertain priors, or multiple uncertain data models. Second, MACE defines regularized version in terms of balance equations rather than optimization. This allows for a much more general framework of well-defined inverse operators with potentially superior performance.

This presentation is part of Minisymposium “MS55 - Advances of regularization techniques in iterative reconstruction (2 parts)
organized by: Zichao (Wendy) Di (Argonne National Lab) , Marc Aurèle Gilles (Cornell University) .

Charles Bouman (Purdue University)