Study of Group Equivariant Convolutional Networks For Image Classification

Image Classification has grabbed a lot of attention in the Artificial Intelligence domain. Convolutional Neural Networks (CNNs) are commonly used for this purpose. A further advancement in CNNs, the Group Equivariant Convolutional Networks or G-CNNs, use G-convolutions, a revolutionary type of neural layer which can share weight more efficiently than conventional convolution layers. This novel approach can make the existing CNN architecture rotation as well as reflection equivariant. To begin, this paper reviews the existing CNN model and areas in which Group Equivariant Convolutional Networks modify this existing architecture. Moreover, this paper also shows the implementation of the G-CNN model on various datasets and conclusions are drawn regarding the efficacy of this architecture.