초록 ▼

A data processing system and method for solving pattern classification problems and function-fitting problems includes a neural network in which N-dimensional input vectors are augmented with at least one element to form an N+j-dimensional projected input vector, whose magnitude is then preferably n

A data processing system and method for solving pattern classification problems and function-fitting problems includes a neural network in which N-dimensional input vectors are augmented with at least one element to form an N+j-dimensional projected input vector, whose magnitude is then preferably normalized to lie on the surface of a hypersphere. Weight vectors of at least a lowest intermediate layer of network nodes are preferably also constrained to lie on the N+j-dimensional surface. To train the network, the system compares network output values with known goal vectors, and an error function (which depends on all weights and threshold values of the intermediate and output nodes) is then minimized. In order to decrease the network\s learning time even further, the weight vectors for the intermediate nodes are initially preferably set equal to known prototypes for the various classes of input vectors. Furthermore, the invention also allows separation of the network into sub-networks, which are then trained individually and later recombined. The network is able to use both hyperspheres and hyperplanes to form decision boundaries, and, indeed, can converge to the one even if it initially assumes the other.

대표청구항 ▼

A data processing system comprising: a) input means for representing each of a series of input data groups as a sequence of N numerical values to form corresponding N-dimensional base input vector, and for storing each base input vector; b) neural network means including: i) input layer storage mean

A data processing system comprising: a) input means for representing each of a series of input data groups as a sequence of N numerical values to form corresponding N-dimensional base input vector, and for storing each base input vector; b) neural network means including: i) input layer storage means, intermediate layer storage means and output layer storage means; ii) said input layer storage means comprising xat least N+j projected input memory units, where j is a predetermined positive integer, for storing a normalized projected input vector having N+j numerical elements, with each projected input vector corresponding to one of the base input vectors; iii) said intermediate layer storage means comprising a plurality of intermediate memory units for storing predetermined intermediate threshold values and intermediate weight vectors; iii) said intermediate layer storage means comprising a plurality of intermediate memory units for storing hidden node network values; iv) said output layer storage means comprising a network output node for storing a network output value; v) connection means for connecting each projected input memory unit with predetermined ones of the intermediate memory units and for connecting the output node with predetermined ones of intermediate memory units; c) processor and control means; i) for augmenting each N-dimensional base input vector with j projection elements to form said projected input vector; ii) for computing an intermediate threshold value and an intermediate weight vector, with each weight vector having N+j weight elements, for each intermediate memory unit in a lowest intermediate layer; and iii) for computing an output value as a predetermined function of the intermediate weight vectors, the intermediate thresholds values and the projected input vectors; d) connection means; i) for connecting the input means to the processor and control means; ii) for connecting the input means to the neural network means; and iii) for connecting the neural network means to the processor and control means.