Equivariant Gaussian Processes as Limiting Convolutional Networks with Infinite Number of Channels
December 01, 2021
January 12, 2022
In this work we establish a relationship between the many-channel limit for $SO(2)$-equivariant convolutional neural networks (CNNs) and the corresponding equivariant Gaussian process (GP) in the case of Fourier space quadratic nonlinearity. The approach used provides explicit equivariance at each step of the relationship derivation.