What's the most efficient way of computing AA'?

autograd
1

If I’m only interested in the upper triangle of B (B will be symmetric), what’s the most efficient way of computing it in PyTorch?

B = torch.bmm(A, torch.transpose(A))

2

Hi,

I don’t think we have anything to do this.
In particular, we have no way to represent a triangular matrix without having the full matrix.
And we have almost no function that takes triangular matrices as input either.