摘要 |
PROBLEM TO BE SOLVED: To efficiently learn dynamics. SOLUTION: A winner node determination part 7-2 determines a winner node that is a node corresponding to the dynamics most suitable for an observed time-serial data, out of the plurality of nodes in a network constituted of the plurality of nodes for holding the dynamics. A learning weight determining part 7-3 determines a weight of learning of the dynamics held by the node in every of the nodes, in response to a distance from the winner node. A parameter updating part 7-4 learns self-organizingly the each dynamics of the network by a degree corresponding to the weight of learning, based on the time-serial data. This learning device/method of the present invention is applicable, for example, for a robot. COPYRIGHT: (C)2008,JPO&INPIT
|