Group Equivariant Convolutional Networks

24 Feb 2016  ·  Taco S. Cohen, Max Welling ·

We introduce Group equivariant Convolutional Neural Networks (G-CNNs), a natural generalization of convolutional neural networks that reduces sample complexity by exploiting symmetries. G-CNNs use G-convolutions, a new type of layer that enjoys a substantially higher degree of weight sharing than regular convolution layers. G-convolutions increase the expressive capacity of the network without increasing the number of parameters. Group convolution layers are easy to use and can be implemented with negligible computational overhead for discrete groups generated by translations, reflections and rotations. G-CNNs achieve state of the art results on CIFAR10 and rotated MNIST.

PDF Abstract

Datasets


Results from Other Papers


Task Dataset Model Metric Name Metric Value Rank Source Paper Compare
Colorectal Gland Segmentation: CRAG G-CNN (C4) F1-score 0.833 # 7
Dice 0.856 # 7
Hausdorff Distance (mm) 170.4 # 8
Multi-tissue Nucleus Segmentation Kumar G-CNN (C4) Dice 0.793 # 14
Hausdorff Distance (mm) 49.0 # 15
Breast Tumour Classification PCam G-CNN (C4) AUC 0.964 # 6

Methods