PottsMGNet: A Mathematical Explanation of Encoder-Decoder Based Neural Networks

Abstract

SIAM Journal on Imaging Sciences, Volume 17, Issue 1, Page 540-594, March 2024.
Abstract. For problems in image processing and many other fields, a large class of effective neural networks has encoder-decoder-based architectures. Although these networks have shown impressive performance, mathematical explanations of their architectures are still underdeveloped. In this paper, we study the encoder-decoder-based network architecture from the algorithmic perspective and provide a mathematical explanation. We use the two-phase Potts model for image segmentation as an example for our explanations. We associate the segmentation problem with a control problem in the continuous setting. Then, the continuous control model is time discretized by an operator-splitting scheme, the PottsMGNet, and space discretized by the multigrid method. We show that the resulting discrete PottsMGNet is equivalent to an encoder-decoder-based network. With minor modifications, it is shown that a number of the popular encoder-decoder-based neural networks are just instances of the proposed PottsMGNet. By incorporating the soft-threshold-dynamics into the PottsMGNet as a regularizer, the PottsMGNet has shown to be robust with the network parameters such as network width and depth and has achieved remarkable performance on datasets with very large noise. In nearly all our experiments, the new network always performs better than or as well as on accuracy and dice score compared to existing networks for image segmentation.