Convolutional Layers

All the convolution layers follow the same pattern. They support all of the functionality of the regular convolution (although not everything is tested).

They all have the additional arguments:

• group (str) - The group with which to apply the convolution.

• subgroup (str, default None) - To use a subgroup to convolve a signal on the full group. Defaults to None, which means that the subgroup is the same as the group.

• allow_non_equivariance (default: False) - whether to allow non-equivariant operations. This is used to implement non-equivariant convolutions.

Furthermore there are additional options for padding:

• padding=’valid_equiv’ - Valid padding with minimal extra padding to ensure equivariance.

• padding=’same_equiv’ - Same padding with minimal extra padding to ensure equivariance.