Generally, routing methods can be implemented using centralized and decentralized methods. In order to improve the efficiency of a telecommunications network, we want a routing method to have the advantages of both centralized routing - optimal decision and decentralized routing - fast and less work load. The neural network routing uses the centralized routing as teacher, such that it can control with only the local information. We study the learning process of the neural network and use the Estimation and Maximization (EM) algorithm for learning. The result shows some improvements of neural network over previous work, and the performance of the improved neural network routing is almost the same as the EM algorithm of one hierarchy. By comparing different routing methods: Simplex Routin...[ Read more ]

Generally, routing methods can be implemented using centralized and decentralized methods. In order to improve the efficiency of a telecommunications network, we want a routing method to have the advantages of both centralized routing - optimal decision and decentralized routing - fast and less work load. The neural network routing uses the centralized routing as teacher, such that it can control with only the local information. We study the learning process of the neural network and use the Estimation and Maximization (EM) algorithm for learning. The result shows some improvements of neural network over previous work, and the performance of the improved neural network routing is almost the same as the EM algorithm of one hierarchy. By comparing different routing methods: Simplex Routing, Maximum Free Circuit Routing, Neural Routing, Free Circuit Ratio Routing, we can see that neural routing is better than typical decentralized routing. Finally, we can see that Free Circuit Ratio Routing have a better performance in a hierarchical network with extensive connectivities.