Beyond PDP: the frequency modulation neural network architecture
Hideto Tomabechi, Hiroaki Kitano · International Joint Conference on Artificial Intelligence · 1989
This paper proposes the Frequency Modulation Neural Network as an alternative to current neural-net models. This proposal is for an architecture of a heterogeneous neural-network in which information is propagated using frequency modulation of pulses oscillated by groups of neurons. The FMNN model enables operations including variable-binding, sequential recognitions and predictions. The use of FM signals for communication among neural clusters also enables the model to avoid communication bottlenecks arising in most massively parallel computer architectures.