Basin of Attraction of Associative Memory as it Evolves from Random Weights
機関リポジトリ (HANDLE)
オープンアクセス
説明
We apply genetic algorithms to fully connected Hopfield associative memory networks. Previously, we reported that a genetic algorithm can evolve networks with random synaptic weights to store some number of patterns by pruning some of its synapses. The associative memory capacity obtained in that experiment was around 16% of the number of neurons. However the size of basin of attraction was rather small compared to the original Hebb-rule associative memory. In this paper, we present a new version of the previous method trying to control the basin size. As far as we know, this is the first attempt to address the size of basin of attraction of associative memory by evolutionary processes.