Transfert d'information d'un réseau connexionniste vers le système cognitif humain par pseudo patterns
Serban C. Musca, Stéphane Rousset, Bernard Ans · HAL (Le Centre pour la Communication Scientifique Directe) · 2001
In the present paper we show that humans are sensitive to distributed information. The distributed information used was conveyed by pseudo-patterns. Pseudo-patterns result from injecting random noise into a connectionist network that has previously learned the information to transmit. As pseudo-pattern is a concept related to the recently proposed solution (Robins, 1995; Ans et Rousset, 1997, 2000; French, 1997) to the problem of catastrophic forgetting in distributed connectionist networks, we briefly review this research field and highlight the relevance of the use of pseudo-patterns in distributed networks. We show that the human brain is sensitive to pseudo-patterns, in the sense that distributed information generated by an external reverberant connectionist system is an efficient learning basis for humans. On this account, we think that new ways are opened in understanding the mechanism(s) of information integration, consolidation and transfer in human memory.