When calculating the pairwise ranking loss, should the ReLU be applied to the element-wise multiplication of two difference matrix (as is in the paper, equation (8)) instead of after the matrix multiplication, where the difference product is already aggreggated along one dimension)? Thanks!
Hi there,
When calculating the pairwise ranking loss, should the ReLU be applied to the element-wise multiplication of two difference matrix (as is in the paper, equation (8)) instead of after the matrix multiplication, where the difference product is already aggreggated along one dimension)? Thanks!