Mon Jan 10 00:10:48 EST 2011
Matrix multiplication is quadratic?
Basic idea: Matrix multiplication can be related to group
multiplication, which can then be implemented by the fourier
This is freaky stuff. I think this paper applies to finite field
matrices though. Does this also work for the reals, or in
approximation for floating point numbers?
Does it mean that in general, _all_ matrix multiplication problems can
be expressed as convolution problems? I.e. the generalized Levinson /
Schur style algorithms already exploit this in some sense.. Are all
matrices "structured" ?