Unsupervised Competitive Neural Networks (UCN) have been recognized as a powerful tool for pattern analysis, feature extraction and clustering analysis. Nevertheless, the inhibitory interactions among the units of the network, required by the winner-take-air paradigm, constitute a crucial step for the implementation of competitive networks in analog VLSI. The aim of this letter is to present an unsupervised competitive neural network characterized by local inhibitory interactions among its cells. The kernel of this network is a neural unit based on a modified competitive teaming law in which the threshold changes in the teaming stage. It is shown that the proposed neuron unit is able, during the learning stage, to perform an automatic selection of patterns that belong to a cluster, moving towards its centroid. The properties of this network, related to the robustness of the final results and to the choice of the number of the elements, are examined in a set of numerical simulations adopting a data set composed of Gaussian mixtures and uniform noise.
Local Competitive Signals for an Unsupervised Competitive Neural Network / Chiarantoni, E.; Acciani, G.; Vacca, F.. - STAMPA. - (2000), pp. 590-593. (Intervento presentato al convegno IEEE International Symposium on Circuits and Systems, ISCAS 2000 tenutosi a Geneva, CH nel May 28-31, 2000) [10.1109/ISCAS.2000.856129].
Local Competitive Signals for an Unsupervised Competitive Neural Network
E. Chiarantoni;G. Acciani;F. Vacca
2000-01-01
Abstract
Unsupervised Competitive Neural Networks (UCN) have been recognized as a powerful tool for pattern analysis, feature extraction and clustering analysis. Nevertheless, the inhibitory interactions among the units of the network, required by the winner-take-air paradigm, constitute a crucial step for the implementation of competitive networks in analog VLSI. The aim of this letter is to present an unsupervised competitive neural network characterized by local inhibitory interactions among its cells. The kernel of this network is a neural unit based on a modified competitive teaming law in which the threshold changes in the teaming stage. It is shown that the proposed neuron unit is able, during the learning stage, to perform an automatic selection of patterns that belong to a cluster, moving towards its centroid. The properties of this network, related to the robustness of the final results and to the choice of the number of the elements, are examined in a set of numerical simulations adopting a data set composed of Gaussian mixtures and uniform noise.I documenti in IRIS sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.
