The USC Andrew and Erna Viterbi School of Engineering USC Signal and Image Processing Institute USC Ming Hsieh Department of Electrical and Computer Engineering University of Southern California

Technical Report USC-SIPI-174

“Computational Study of Structured Networks for Linear Algebra”

by Chung-Kuang Chu and Jerry M. Mendel

April 1991

Structured networks based on the error back-propagation method can solve many linear algebra problems. In this report, the computational requirements for structured networks are analyzed in order to provide computation time information. With this data, we can roughly evaluate the performance of structured networks.

First, computation counts are estimated for serial implementations. These counts are based on a computation schedule which is provided for each linear algebra problem. We also give results for the theoretical processing time (which is estimated from the computation counts) and the actual processing time. Finally, we estimate the processing time for parallel implementations by introducing the basic computing units, which perform the inner product operations. Computation schedules are also provided for the parallel implementations.

To download the report in PDF format click here: USC-SIPI-174.pdf (1.0Mb)