Dynamics & viability of an adaptive network, and comparison between neural and social networks


This report aims at studying the contribution of the viability theory in the understanding of dynamics underlying self-organization among connexionist networks. We introduce control criteria ensuring the survival of the network taken as an adaptive system, in a dynamic view and by focusing on the coordination of the agents and on new behavior tending to emerge from their collaboration. The study reviews both neural and social networks. We first detail the way a neural network can organize itself and structurally evolve in order to satisfy sensori-motor constraints (reinforcement learning, autonomous adaptation, coalitions of strongly connected synchronous neurones). We then review how this formalism can apply to social networks of economic or cultural production. In particular, we try to understand evolution-guiding mechanisms that make agents and cells work better together than individually (collective intelligence) and how the integration is finally being achieved by the network.

full text (in French): (dvi (FR)) (pdf (FR))


Last modified: Sun Dec 18 13:14:02 CET 2005