Self-adaptive multidimensional Euclidean neural networks for pattern recognition

Summary form only given, as follows. A novel family of neural networks and learning algorithms is introduced: the self-adaptive Euclidean networks. At the same time an interpretation of these neural networks as hyperpolyhedra in the N-dimensional Euclidean space is suggested. These networks can self-adapt to a continually changing environment by properly changing the orientation of the faces of a hyperpolyhedron as well as its volume. More than one hyperpolyhedron may be used at the same time. At any time the current structure of each hyperpolyhedron reflects the structure of the current outside world. The network optimally classifies its noise-distorted excitations into categories after a competition between all possible categories. The most frequently used categories are the ´conscious´ ones, and the least used are the ´subconscious´ ones. New categories can be created, and the old ones can be changed, or forgotten if they are not used for a ´long´ time. This neural network can be used to make high-level decisions concerning the environment to which it is exposed.<>