Analysis and Implementation of Simple Dynamic Binary Neural Networks

摘要

This paper presents simple dynamic binary neural networks characterized by ternary connection parameters and signum activation function. The network is based on the 3-1 binary neuron that converts three inputs to one output. Depending on the connection parameters, the network can generate various binary periodic orbits. The networks bring benefits to network hardware and to computation costs in numerical analysis. Using simple feature quantities, stability of target periodic orbits is analyzed. Using simple majority decision circuits, a simple circuit hardware is designed. Performing Verilog simulation, the circuit operation is investigated.