Groups
Category
Equivariance means that applying a transformation before a function is the same as applying a corresponding transformation after the function.
Weight initialization sets the starting values of neural network parameters so signals and gradients neither explode nor vanish as they pass through layers.