Statistical Analysis of Chaotic Neurons in Mutully-Connected Chaotic Neural Network for Solving Quadratic Assignment Problem(Researches and Overseas Activities)--(Research Reports)
松浦, 隆文Takafumi, Matsuura
日本工業大学研究報告 = Report of researches, Nippon Institute of Technology
65 , 2017-09
The quadratic assignment problem (QAP) is one of the most famous combinatorial optimization problems which belong to NP-hard. To solve the QAP, a method which uses mutually-connected chaotic neural network (CNN) has already been proposed. In the method, chaotic dynamics of the CNN effectively controls to avoid the local minima and to search optimal or near-optimal solutions. However, it is not so easy to generate feasible solutions from the CNN, because an output of a chaotic neuron takes an analog value. To generate a feasible solution from the CNN, a solution decision method has already been proposed. In this paper, to generate a better solution from the CNN, we analyze an inter spike interval of the chaotic neurons by using statistical measures such as a coefficient of variation and a local variation, which are frequently used in the field of neuroscience.