Transputer-based parallel systems for performance evaluation of bidirectional associative memory

Parallel implementation of an artificial neural network for pattern association, the bidirectional associative memory (BAM), is discussed. Transputer-based parallel architectures like the hypercube, mesh and linear array were used to exploit parallelism in the BAM. A comparative study of utilization and speedup for various architectures was made. The hypercube performed better in terms of utilization and speedup when compared to the other architectures.<
>