Cresceptron: A Self-organizing Neural Network Which Grows Adaptively

Cresceptron represents a new approach t o neural networks. It uses a hierarchical f m m e w o r k t o grow neural networks automatically, adaptively and incrementally through learning. A t every level of the hierarchy, new concepts are detected automatically and the network grows by creating new neurons and synapses which memorize the new concepts and their context. The training samples are generalized t o other perceptually equivalent i t ems through hierarchical tolerance of deviation. The neural network recognizes the learned i tems and their variations by hierarchically associating the learned knowledge with the input . I t segments the recognized i tems f r o m the input through back tracking along the response paths.