Concept learning: Hierarchical system
Larry Venetsky · 2002
A hierarchical learning system was designed and simulated. The principal investigative tool was a perception-driven, goal-oriented control system. The system utilizes a multilayered neural network with a backpropagation learning mechanism, a set of competitive networks for feature extraction, and a set of neuron layers for performing XOR, OR, and AND operations. The author examines (a) conceptual learning (CL), that is, generating a complete set of Horn clauses with subsequent generalization, and (b) quantitative learning (QL), that is, adjusting the strength of connections (synapses) between nodes in a neural network.>