資源描述:
《Artificial Neural Networks人工神經(jīng)網(wǎng)絡(luò).ppt》由會(huì)員上傳分享,免費(fèi)在線閱讀,更多相關(guān)內(nèi)容在教育資源-天天文庫(kù)。
1、ArtificialNeuralNetworks人工神經(jīng)網(wǎng)絡(luò)IntroductionTableofContentsIntroductiontoANNsTaxonomyFeaturesLearningApplicationsISupervisedANNsExamplesApplicationsFurthertopicsIIUnsupervisedANNsExamplesApplicationsFurthertopicsIII15/09/20212ArtificialNeuralNetworks-IContents-IIntroductiontoANNsProcessingelemen
2、ts(neurons)ArchitectureFunctionalTaxonomyofANNsStructuralTaxonomyofANNsFeaturesLearningParadigmsApplications15/09/20213ArtificialNeuralNetworks-ITheBiologicalNeuron10billionneuronsinhumanbrainSummationofinputstimuliSpatial(signals)Temporal(pulses)ThresholdovercomposedinputsConstantfiringstreng
3、thbillionsynapsesinhumanbrainChemicaltransmissionandmodulationofsignalsInhibitorysynapsesExcitatorysynapses15/09/20214ArtificialNeuralNetworks-IBiologicalNeuralNetworks10,000synapsesperneuronComputationalpower=connectivityPlasticitynewconnections(?)strengthofconnectionsmodified15/09/20215Artif
4、icialNeuralNetworks-INeuralDynamicsRefractorytimeActionpotentialActionpotential≈100mVActivationthreshold≈20-30mVRestpotential≈-65mVSpiketime≈1-2msRefractorytime≈10-20ms15/09/20216ArtificialNeuralNetworks-I神經(jīng)網(wǎng)絡(luò)的復(fù)雜性神經(jīng)網(wǎng)路的復(fù)雜多樣,不僅在于神經(jīng)元和突觸的數(shù)量大、組合方式復(fù)雜和聯(lián)系廣泛,還在于突觸傳遞的機(jī)制復(fù)雜?,F(xiàn)在已經(jīng)發(fā)現(xiàn)和闡明的突觸傳遞機(jī)制有:突觸后興奮,突觸后抑制,突
5、觸前抑制,突觸前興奮,以及“遠(yuǎn)程”抑制等等。在突觸傳遞機(jī)制中,釋放神經(jīng)遞質(zhì)是實(shí)現(xiàn)突觸傳遞機(jī)能的中心環(huán)節(jié),而不同的神經(jīng)遞質(zhì)有著不同的作用性質(zhì)和特點(diǎn)15/09/20217ArtificialNeuralNetworks-I神經(jīng)網(wǎng)絡(luò)的研究神經(jīng)系統(tǒng)活動(dòng),不論是感覺、運(yùn)動(dòng),還是腦的高級(jí)功能(如學(xué)習(xí)、記憶、情緒等)都有整體上的表現(xiàn),面對(duì)這種表現(xiàn)的神經(jīng)基礎(chǔ)和機(jī)理的分析不可避免地會(huì)涉及各種層次。這些不同層次的研究互相啟示,互相推動(dòng)。在低層次(細(xì)胞、分子水平)上的工作為較高層次的觀察提供分析的基礎(chǔ),而較高層次的觀察又有助于引導(dǎo)低層次工作的方向和體現(xiàn)其功能意義。既有物理的、化學(xué)的、生理的、心理的分門別
6、類研究,又有綜合研究。15/09/20218ArtificialNeuralNetworks-ITheArtificialNeuronStimulusurest=restingpotentialxj(t)=outputofneuronjattimetwij=connectionstrengthbetweenneuroniandneuronju(t)=totalstimulusattimetyi(t)x1(t)x2(t)x5(t)x3(t)x4(t)wi1wi3wi2wi4wi5NeuroniResponse15/09/20219ArtificialNeuralNetworks-IA
7、rtificialNeuralModelsMcCullochPitts-typeNeurons(static)Digitalneurons:activationstateinterpretation(snapshotofthesystemeachtimeaunitfires)Analogneurons:firingrateinterpretation(activationofunitsequaltofiringrate)Activationofneuronsencod