Inverse Evolution Layers: Physics-informed Regularizers for Deep Neural Networks

TL;DR

This paper introduces inverse evolution layers (IELs), a novel regularization technique inspired by PDE models, to improve neural network outputs by amplifying undesired properties and enforcing shape and noise robustness.

Contribution

The paper proposes inverse evolution layers (IELs) as a new regularization method that integrates PDE-inspired reverse processes into neural networks for enhanced control and robustness.

Findings

IELs effectively mitigate noisy label effects in semantic segmentation.

Curve-motion IELs enforce convex shape regularization, reducing concave outputs.

Theoretical analysis supports IELs as a robust regularization mechanism.

Abstract

Traditional image processing methods employing partial differential equations (PDEs) offer a multitude of meaningful regularizers, along with valuable theoretical foundations for a wide range of image-related tasks. This makes their integration into neural networks a promising avenue. In this paper, we introduce a novel regularization approach inspired by the reverse process of PDE-based evolution models. Specifically, we propose inverse evolution layers (IELs), which serve as bad property amplifiers to penalize neural networks of which outputs have undesired characteristics. Using IELs, one can achieve specific regularization objectives and endow neural networks' outputs with corresponding properties of the PDE models. Our experiments, focusing on semantic segmentation tasks using heat-diffusion IELs, demonstrate their effectiveness in mitigating noisy label effects. Additionally, we develop curve-motion IELs to enforce convex shape regularization in neural network-based segmentation models for preventing the generation of concave outputs. Theoretical analysis confirms the efficacy of IELs as an effective regularization mechanism, particularly in handling training with label issues.