New work on how to construct neural network layers on composite objects of scalars, vectors, bivectors, … --> multivectors! Via Clifford algebras, we generalize convolution and Fourier transforms to multivectors, especially relevant for PDE modeling: