Leveraging differential k-forms in Rn for efficient and interpretable geometric representation learning without message passing.
