Multiplex Communication with Synchronous Shift and Weight Learning in 2D Mesh Neural Network
Takuya Kamimura · Automation Control and Intelligent Systems · 2015
We have previously proposed a multiplex communication system in a neural network. However, this system is designed to force the network to communicate in a multiplexed manner, in which “codes” or “temporal sequences” are inevitably induced. This means that the network has a main loop and coding/decoding circuits, which are somewhat artificial. In this paper, we show that it is also possible to communicate without these artificial guidance aids by multiplexing in a 2D mesh-type neural network, where learning procedures are used to find paths from an originating neuron to a destination neuron. We also provide statistics from these neural networks to show that random sequences occur more frequently than non-random sequences.