For sure, as you point out, many such kernels have been written, going at least as far back as 1967.

We believe our contribution is being able to generate kernels for all the expressions.

Thanks for the reference though! We’re trying to learn where sparsity may be important in neural networks.