イナザワ　ヒロシ   INAZAWA Hiroshi
  稲澤　弘志
   所属   人間科学部 都市生活学科
   職種   教授
言語種別 英語
発行・発表の年月 2018
形態種別 研究論文（学術雑誌）
標題 Hetero-Correlation-Associative Memory with Trigger Neurons: Accumulation of Memory through Additional Learning in Neural Networks
執筆形態 単著
掲載誌名 Complex Systems
掲載区分国外
出版社・発行元 Complex Systems Publications, Inc
巻・号・頁 Volume 27(2),pp.187-197
総ページ数 11
著者・共著者 Hiroshi Inazawa
概要 In this paper, we propose an hetero-correkation associative memory. In the conventional model, it is one problem that the memory rate does not become large. In this model, it is possible to increase the memory rate (the number of neurons used / the number of stored images (image) × 100) in principle up to nearly 100% (not 100%). The proposed model consists of a layer of trigger neurons (T-net) and a neuron layer (M-net) that memorizes. Here, one neuron of T-net corresponds to the storage of one image of M-net. When memorizing a new image, one trigger neuron is added and learning (adjustment of synapse load) is performed with M-net. Therefore, the synaptic weights of T-net neurons and M-net neurons do not interfere with those of other memorized images. This makes it possible to improve the memory rate of this model.
ISSN 0891-2513