Quantitative Comparison of Architectures for Digital Neuro-Computers

for Digital Neuro-Computers Paolo Ienne LAMI & MANTRA, Swiss Federal Institute of Technology, CH-1015 Lausanne In Proceedings of the International Joint Conference on Neural Networks, volume II, pages 1987{90, Nagoya, Japan, October 1993. Abstract As the eld of neural-networks matures toward real-world applications, a strong need for hardware systems to e ciently compute larger network arises. Comparisons between neural-networks dedicated platforms are often fragmentary or qualitative. An architectural level evaluation scheme is attempted by proposing some performance indexes on idealized architecture classes. Some conclusions on the advantages and limitations of the architectures are drawn.