Can you provide a source? For example, the SVD (Golub-Kahan-Reinsch) method generally involves bidiagonalisation of the matrix and then implicit QR steps, whic

Can you provide a source? For example, the SVD (Golub-Kahan-Reinsch) method generally involves bidiagonalisation of the matrix and then implicit QR steps, which are often achieved with Householder reflections and Givens rotations. Sure, all of those are conceptually matrix multiplications, but they're not implemented as O(N^3) matrix multiplications; rather, their special structure is exploited so that they're faster. Yet, the entire algorithm is still cubic. So not sure Strassen would accelerate things.