Reuse of one learnable parameter in multiple computational positions. In a convolution this reduces parameter count and imposes translation equivariance away from boundary effects.
New to topics? Read the docs here!
New to topics? Read the docs here!