“The Categorical Deep Learning framework generalizes Geometric Deep Learning, which is recovered as the special case where the homomorphism corresponds to a group action.”