• What I want

I want to do element-wise matrix multiplication of two matrix:

1. sparse matrix A with shape (3,4,5)

2. dense matrix B with shape (100,10,4,5)

I am expecting to do element-wise matrix multiplication on the last 2 dimensions of these matrix

So the resulting matrix C=A*B have shape (100,10,3,4,5)

I want to do this more efficiently. What can I do?

• What I currently do

I checked the docment of the Pytorch.sparse, which indicates to use the operator * to do element-wise matrix multiplication.

However, this operator only support 2-dimension matrix. I have to iterate over the first dimension of matrix A and the first two dimensions of matrix B to do 2d*2d matrix multiplication to get the right results, which is not efficient.