It's possible to perform a sandwich product of object x with transform y with either the unnormalized form y' * x * y or the normalized form inv(y) * x * y. This should already be pretty fast considering the speed of the multiplication kernel, but I wonder if we can speed this up with a dedicated function.
I also wonder if we can use a module-specific eval() definition to automatically convert manually-implemented sandwich products to a fast kernel.
It's possible to perform a sandwich product of object
xwith transformywith either the unnormalized formy' * x * yor the normalized forminv(y) * x * y. This should already be pretty fast considering the speed of the multiplication kernel, but I wonder if we can speed this up with a dedicated function.I also wonder if we can use a module-specific
eval()definition to automatically convert manually-implemented sandwich products to a fast kernel.