エネルギ-減少への寄与の大きさの順にニュ-ロンを遷移させる機能を有する2値ホップフィ-ルドネット

概要

論文の詳細を見る
In the ordinary binary Hopfield's Net, the neuron selected by random number changes its state asynchronously (We call this the ''random method''). The performance of this network is very low. In order to improve this low convergence, we change the order of application of Hopfeild's transition rule. In our method, the neuron changes its state in descending order of the contribution of neuron to energy decrease (We call this the ''ordered method''). By several numerical simulations it is verified that the convergency in the ordered method is higher than that in the random method in almost all cases. In order to realize this new method we need to compute the amount of energy decrease resulting from the state transition. But the amount of this additional computation is negligiblly small if we compute the difference in the energy decrease resulting from the transition of each neuron.
鹿児島大学,カゴシマダイガク,Kagoshima Universityの論文