A Modified Pulse Coupled Neural Network for Shortest Path Computation

Shortest path computation is a classical combinatorial optimization problem. Neural networks, as a class of effective optimization approaches, have been used for processing path optimization problems for a long time. Pulse Coupled Neural Network (PCNN) is a very different neural network, which has also been proposed to compute shortest paths in recent years. In some existing PCNN models, the computational complexity is closely related to the length of actual shortest path. As a result, these PCNNs suffer from high computational cost for very long paths. This paper proposed a new PCNN model, called Dual Source PCNN (DSPCNN), which can improve the computational efficiency of PCNNs for shortest path problems. Two Autowaves produced by the DSPCNN: one comes from the Source neuron, and the other comes from the Goal neuron. Once the Autowaves from the two firing sources meet, the DSPCNN stops, and then the shortest path is found by backtracking the two Autowaves. Experimental results show that the DSPCNN is more efficient than some existing PCNNs.