ognition,machinevision,nonlinearoptimization,automatictargetidentification,knowledgeproceing,remoteseing,etc.ithasbecomenotonlythetastesofscientistsbutalsotheinterestsofgovernmentsandforces.thegovernmentsandindustrialcommunitiesofmanycountries/regioaresokeenonneuralcomputingtechniquesthattheyhaveinvestedalargeamountofmoneyoncorreondingresearch.thereforetheprogreofneuralcomputingwillnotonlypromotethedevelopmentofscienceandtechnologybutalsoinfluencethenationalpowers.
inthisdiertation,5problemssta
ndinginneedofsolutioareinvestigated,whichincludesexpeditingthelearningeedofneuralnetworks,improvingthecompreheibilityofneuralnetworks,designingengineeringneuralcomputingmethodsthatareeasytouse,simulatingbiologicalneuralsystemsmorebetterthanever,andcombiningneuralcomputingwithtraditionalartificialintelligencetechniques.themaincontributioofthisdiertationaresummarizedasfollows:
firstly,afastneuralclaifiernamedfacandafastneuralregreionestimatornamedfanreareproposed.experimentalresultsshowthatthosetwoalgorithmsthathavefastlearningabilityandstronggeneralizationabilityreectivelyoutperformsomeprevailingneuralclaificationalgorithmsandneuralregreionestimationalgorithmsatpresent.besides,facissuccefullyaliedinlithologyidentificationofoilexploration.
secondly,aneuralnetworkruleextractionalgorithmnamedstareisproposed.experimentalresultsshowthatstarecanextractaccurateandcompactsymbolicrulesthathavehighfidelity,sothatthecompreheibilityoftrainedneuralnetworksareimproved.additionally,