Classification with Runge-Kutta networks and feature space augmentation

TL;DR

This paper introduces a novel neural network architecture that combines Runge-Kutta methods with feature space augmentation, leading to improved numerical performance in point and image classification tasks.

Contribution

It proposes a new network design integrating Runge-Kutta schemes and input space augmentation, enhancing classification accuracy over existing methods.

Findings

Improved classification performance demonstrated on multiple datasets.

Effective integration of Runge-Kutta methods with feature augmentation.

Implementation in PyTorch confirms practical applicability.

Abstract

In this paper we combine an approach based on Runge-Kutta Nets considered in [Benning et al., J. Comput. Dynamics, 9, 2019] and a technique on augmenting the input space in [Dupont et al., NeurIPS, 2019] to obtain network architectures which show a better numerical performance for deep neural networks in point and image classification problems. The approach is illustrated with several examples implemented in PyTorch.