Algorithmic Mapping of Feedforward Neural Networks onto Multiple Bus Systems

• Authors: Ahmed El-Amawy;Priyalal Kulasinghe
• Venue: IEEE Transactions on Parallel and Distributed Systems
• Year: 1997

This paper addresses the problem of mapping a feedforward ANN onto a multiple bus system, MBS, with p processors and b buses so as to minimize the total execution time. We present an algorithm which assigns the nodes of a given computational layer (c-layer) to processors such that the computation lower bound $\lceil N^l /p \rceil t_p^l$ and the communication lower bound $\lceil N^l/b \rceil t_c$ are achieved simultaneously, where $N^l$ is the number of nodes in the mapped c-layer $\ell$ and $t_p^l$ and tc are the computation and communication times, respectively, associated with a node in the layer. When computation and communication are not overlapped, we show that the optimal number of processors needed is either 1 or p, depending on the ratio $t_p^l /t_c.$ When computation and communication are overlapped, we show that the optimal number of processors needed is either 1 or $(\lceil t_p^l / t_c \rceil ) b.$ We show that there is a unique arrangement of interfaces such that the total number of interfaces is minimum and the optimal time is reached. Finally, we compare the relative merits of the MBS simulating ANNs over the recently introduced checkerboarding scheme.