On stability and associative recall of memories in attractor neural networks.

摘要: Attractor neural networks such as the Hopfield model can be used to associative memory. An efficient memory should able store a large number of patterns which must all stable. We study in detail meaning and definition stability network states. reexamine meanings retrieval, recognition recall assign precise mathematical each these terms. also examine relation between them how they relate capacity network. have shown earlier this journal that orthogonalization scheme provides an effective way overcoming catastrophic interference limits model. It is not immediately apparent whether improvement made by orthgonalization affects processes equally. show influence occurs different degrees hence relations them. then conditions for pattern split into necessary condition (recognition) sufficient one (recall). interpret cognitive terms information being stored after it orthogonalized. alterations dynamics upon introduction orthogonalization, their effects on efficiency