Um Método de Aprendizado para Redes Neuronais Analógicas de Hopfield com Aplicacão

of Thesis presentecl to COPPE as partia1 fulfillinent of tlie requireinents for tlie degree of Master of Science ( R 4 Sc.) A Learning Method for Neural Aiialog Hopfield Networlís with application in Image Coinpression Jaiie Tavares Alvarez Deceinber, 1992 Thesis Supervisors: Luis Alfreclo Viclal de Casvallio Valinir C. Ba.rbosa. Department: Programa de Engeilhasia de Sistemas e Coinputação. 111 this worlí, three new algoritl-iins for Ailalog Hopfield Neural Networlts are presentecl, enlpirically evaluatecl with rancloin patterns. The results obtained iu leariiing overcame the oiies which liave been ltiiown. Once cletermined tlie algorithm of better perforinaiice, we begail the application tests vr~liicli are coinposect of two stages: the learning of iinages of 400 pixels in black , gray ancl wliite colors ailcl the coinpressioil (accomplished by ai1 intuitive proposecl method) followiiig by tlie clecoinpressioil (accoinplislied by a Hopfielcl networlí) of these images. It is interesting to notice tliat iil tlie first stage of the application tests, a11 tl-ie images groups liave been 100% learneci. Tl-ie seconcl stage also proclucecl goocl results, mainly if we consicler the visual aspect of the images. A11 the tests have been perforined 011 a worlistatioii (SUN), utilizing tl-ie C language as prograinming tool.