A self adaptive object oriented implementation of an analog ARTMAP neural network
Ian Taylor, Mike Greenhough · 2002
This paper describes the implementation of a self-adaptive object-oriented analog ARTMAP algorithm. The ARTMAP network consists of two self-adaptive ART 2-A networks for the ART/sub a/ and ART/sub b/ networks, which are connected by a self-adjusting map-field network. The self-adaptive ART 2-A networks allow automatic re-adjustment of their /spl Fscr//sub 2/ layers by dynamically allocating /spl Fscr//sub 2/-node objects as and when they are required. The map-field network then adjust itself to accommodate the ever-growing sizes of the two /spl Fscr//sub 2/ layers of the ART 2-A networks. This self-adaptive mechanism allows the network to maintain a high degree of self-organisation, that is, the user does not need to set pre-conditions about the size of each of the network's /spl Fscr//sub 2/ layers (or the map-field) when applying it to a new problem domain.