Re:Limits on Matrix Multiplication
by bodas on Saturday September 01, 2018 @08:07AM
(#17891446)
Deep learning is done in hardware so you're not optimising instructions but circuits.
[ Parent
| on HN ]
Not sure I understand this. The circuit for addition is easier than the circuitry for multiplication.
That said, in circuits, it is far less likely to ever multiply that large of a matrix. Even if I were correct that many steps of large training can be cast as a matrix multiply.