Gauss-Newton Natural Gradient Descent for Shape Learning

TL;DR

This paper introduces a Gauss-Newton based optimization method for shape learning that improves convergence speed and stability over traditional methods, demonstrated through experiments on benchmark tasks.

Contribution

The paper presents a novel application of the Gauss-Newton method to shape learning, addressing ill-conditioning and convergence issues in neural surface optimization.

Findings

Faster convergence compared to first-order methods

More stable optimization process

Improved final solution accuracy

Abstract

We explore the use of the Gauss-Newton method for optimization in shape learning, including implicit neural surfaces and geometry-informed neural networks. The method addresses key challenges in shape learning, such as the ill-conditioning of the underlying differential constraints and the mismatch between the optimization problem in parameter space and the function space where the problem is naturally posed. This leads to significantly faster and more stable convergence than standard first-order methods, while also requiring far fewer iterations. Experiments across benchmark shape optimization tasks demonstrate that the Gauss-Newton method consistently improves both training speed and final solution accuracy.