Advanced search


Juan C. Pérez-Cortés, Enrique Vidal. GSLA: A Geometric Supervised Learning Algorithm. Proceedings of Neuro-Nimes 91, Neural Networks and their Applications, 1991. pp. 253-268.

A method is presented that, while using classical Pattern Recognition (PR) concepts and techniques, behaves as a black box similarly to a multilayer perceptron. Although some architectural peculiarities and advantages of neural networks are not fully achieved by the algorithm, others such massive parallelism, failure tolerance and the possibility of analog harware implementation are features that both approaches share. Based on classical unsupervised clustering techniques, the proposed approach takes advantage of a supervised training set of input-output data, to learn a piecewise-linear mapping between an input and an output space. Experimental evidence is presented showing excellent learning speed and robustness.