Takayuki Kimura; Hiroyuki Nakajima; Tohru Ikeguchi
PHYSICA A-STATISTICAL MECHANICS AND ITS APPLICATIONS 376 658 - 672 2007年03月
[査読有り] We propose a new algorithm for controling packet routing by a neural network. First, we show that the conventional method for a packet routing control using a neural network is not so effective when it is applied to a network with irregular topology. To overcome this problem, we propose a modified method with stochastic effects. In the present paper, we evaluated our proposed method for irregular, randomized and scale-free networks. Further, we compared our method with the conventional method and another modified method. We show that our proposed method exhibits better performance than the conventional routing method.
Introducing some additional measures, we also analyze why the proposed method shows better performance. As a result, we confirmed that our proposed method skillfully decentralizes the packets in the computer network, and which improves the performance of the proposed method. (c) 2006 Elsevier B.V. All rights reserved.