β Scribed by Toshio Tanaka; Tetsuya Higuchi; Tatsumi Furuya
No coin nor oath required. For personal study only.
When solving optimization problems on Hopfield neural networks, good solutions are not obtained due to convergence to local minima of the energy function. The Boltzmann machine can escape from local minima because of its stochastic behavior, but the computation time is very long to reach the (semiβ) global minimum.
To avoid these problems, this paper proposes an efficient algorithm for solving optimization problems by the use of binary Hopfield neural networks. This algorithm selects a neuron which reduces the energy function value the most to reduce the energy of the network efficiently. Those algorithms are applied to the traveling salesman problem (TSP) with 10 cities. Nearβoptimal solutions were obtained more efficiently than an ordinary TSP algorithm because the energy level of the network can reach lower energy states. The computation time was reduced by one order of magnitude because the energy level of the network reached lower energy states much faster than an ordinary TSP algorithm.
π SIMILAR VOLUMES