Exponential and chaotic neurodynamical tabu searches for quadratic assignment problems

We propose a chaotic neurodynamical searching method for the Quadratic Assignment Problems (QAPs). First, we construct a neural network whose behavior is the same as that of the conventional tabu search. Using the dynamics of the tabu search neural network, we realize the exponential tabu search, whose tabu effect decreases exponentially with time, and we show the effectiveness of this type of exponential tabu search. Next, we extend this novel tabu search to a chaotic version. This chaotic method includes both effects of the chaotic dynamical search and the exponential tabu search, and exhibits better performance than the conventional and exponential tabu searches. Last, we propose an automatic parameter tuning method and show that the proposed method exhibits high performance even on large QAPs.