Given two real values matrices A and B with size m x n and n x l, I'm interested in the sign of the elements in their matrix multiplication, i.e.,:
C = sign(A x B)
Given the fact that both A and B are very large matrices and I'm only interested in the sign of the output elements, is there a way to approximate C rather than computing this computationally expensive matrix multiplication?
I'm looking for an approximation and a certain degree of error would be totally acceptable.
PS: One common scenario is when the elements of A and B are -1 or 1