Mon. 4:15pm tutorial: small correction regarding commutativity of matrix multiplication

Disclaimer: Dieser Thread wurde aus dem alten Forum importiert. Daher werden eventuell nicht alle Formatierungen richtig angezeigt. Der ursprüngliche Thread beginnt im zweiten Post dieses Threads.

Mon. 4:15pm tutorial: small correction regarding commutativity of matrix multiplication
Dear all,

during the tutorial I tentatively agreed that the matrix multiplication can be re-ordered for more efficient computation.
I thought a bit about it, and don’t think it’s true. If you think conceptually about it, this would mean that the order of the evidence doesn’t play a role anymore.

From a mathematical point of view, AB = BA works if both A and B are diagonal matrices. However, in our case, only the sensor model matrices were diagonal, but the transition matrices were not – and the reordering was illegal.


FWIW you can, however, regroup multiple multiplicands, e.g. A*(BC) to (AB)*C, and that might very well give you performance boosts :slight_smile: