Systolic algorithm for tensor products of matrices: Implementation and applications (Q915381)

A systolic algorithm is presented for computing the tensor T of r matrices each of size \(n\times n\) and multiply the result by an \(n^ r\) vector D. A so-called ``bullet-l'' higher dimensional operator is introduced which consists of \(n^{r-1}\) matrix\(\times vector\) multiplications. This operator leads to a recursive calculation of \(T\cdot D\). A systolic implementation of the bullet operation is discussed and a hypercube implementation of the tensor algorithm is presented. The resulting computing time is given for each of these implementations. Some applications to interpolation and approximation of functions as well as graphics and image processing are mentioned.