International Symposium on Nonlinear Theory and its Applications
Dynamical Noise injection to Chaotic Dynamics for Solving Combinatorial Optimization Problems
Takayuki Suzuki, Takafumi Matsuura, Tohru Ikeguchi,
PDF download (57.1KB)
To solve the quadratic assignment problems (QAPs), two types of chaotic search methods have already been proposed. In one method, mutual connection chaotic neural network (CNN), Hopfield-type CNN method is used, and a firing pattern of the CNN represents a solution of the QAP. For another method, execution of local search algorithm is control by the chaotic dynamics. In both methods, chaotic dynamics works to avoid local minima. To improve performances of the these methods, we have already proposed new methods which combine chaotic dynamics and dynamical noise. As a result, when small amount of dynamical noise is added to the CNN, the solving performance is improved. However, we have not clarified yet why the small amount of dynamical noise is effective to find good solutions and how to change the searching states of the chaotic neural network by the dynamical noise. In this paper, to clarify the reason, we analyze the internal states of the neurons.