Exploiting Cyclic Symmetry in Convolutional Neural Networks

TL;DR

This paper introduces four novel operations that can be integrated into neural networks to encode rotational symmetry, enabling models to be partially equivariant to rotations and improving performance on symmetric datasets.

Contribution

The authors propose four new layer operations that encode rotational symmetry in neural networks, reducing the need for data augmentation and enhancing efficiency.

Findings

Improved accuracy on datasets with rotational symmetry

Smaller models achieve comparable or better performance

Parameter sharing across orientations enhances efficiency

Abstract

Many classes of images exhibit rotational symmetry. Convolutional neural networks are sometimes trained using data augmentation to exploit this, but they are still required to learn the rotation equivariance properties from the data. Encoding these properties into the network architecture, as we are already used to doing for translation equivariance by using convolutional layers, could result in a more efficient use of the parameter budget by relieving the model from learning them. We introduce four operations which can be inserted into neural network models as layers, and which can be combined to make these models partially equivariant to rotations. They also enable parameter sharing across different orientations. We evaluate the effect of these architectural modifications on three datasets which exhibit rotational symmetry and demonstrate improved performance with smaller models.