International Symposium on Nonlinear Theory and its Applications
DT-CNN Annealing with Additive Noise Generated by Class 3 CA: A Comparison with Chaos Annealing
Tomohiro FUJITA, Takeshi OGURA,
PDF download (302.8KB)
An annealing for global optimization on discrete time cellular neural network (DT-CNN) with additive noise is investigated. In our method the additive noise is generated using chaotic behavior of class 3 cellular automata (CA). The optimization with our method shows higher performance than with uniform random noise. In this paper we investigate the relation between our method and chaos annealing, which is another superior optimization method with neural network. The auto-correlations of the CA noise and logistic map, which is often used in chaos annealing, are evaluated, and we can see the similarity of both systems. The annealing performances are also evaluated, and the results with the CA noise and logistic map show both methods exhibit higher performances under the same condition. These results may suggest that the CA noise and logistic map have common basis and it makes the ability of global search high.