Learning by choice of internal representations

摘要: We introduce a learning algorithm for multilayer neural networks composed of binary linear threshold elements. Whereas existing algorithms reduce the process to minimizing cost function over weights, our method treats internal representations as fundamental entities be determined. Once correct set is arrived at, weights are found by local and biologically plausible Perceptron Learning Rule (PLR). tested on four problems: adjacency, symmetry, parity combined symmetry-parity.