是训练样本的个数

《是训练样本的个数》由会员上传分享，免费在线阅读，更多相关内容在行业资料-天天文库。

1、gradientdescentSearchinErrorSpaceFindingtheminimumoffunction:gradientdescentWedefinetheerrorfunction(Meansquareerror,MSE)N是訓練樣本的個數,t是期望的output,y是網路實際的output,我們想找出一組weightW使得E值最小Findingtheminimumoffunction:gradientdescentWeputWhereisthelearningrateBackpropag

2、ation反饋學習規則ConvergenceNeuralnetworkanalysisoftenrequiresalargenumberofindividualrunstodeterminethebestsolution.Mostlearningruleshavebuilt-inmathematicaltermstoassistinthisprocesswhichcontrolthe'speed'(Beta-coefficient)andthe'momentum'ofthelearning.Convergenc

3、eThespeedoflearningisactuallytherateofconvergencebetweenthecurrentsolutionandtheglobalminimum.Momentumhelpsthenetworktoovercomeobstacles(localminima)intheerrorsurfaceandsettledownatorneartheglobalminimum.GeneticAlgorithmsMotivationLookatwhatevolutionbringsus

4、?VisionHearingSmellingTasteTouchLearningandreasoningCanweemulatetheevolutionaryprocesswithtoday'sfastcomputers?7/24/20218GeneticAlgorithmsTerminology:FitnessfunctionPolulationEncodingschemesSelectionCrossoverMutationElitism7/24/20219GeneticAlgorithmsBinaryen

5、coding(11,6,9)101101101001GeneChromosomeCrossoverMutation10011110101100101011111010010010Crossoverpoint1001111010011010Mutationbit7/24/202110Step1:Establishabasepopulationofchromosomes.Step2:Determinethefitnessvalueofeachchromosome.Step3:Duplicatethechromoso

6、mesaccordingtotheirfitnessvaluesandcreatenewchromosomesbymatingcurrentchromosomes.(e.g.,mutation,recombination).Step4:Deleteundesirablemembersofthepopulation.Step5:Insertthenewchromosomesintothepopulationtoformanewpopulationpool.GeneticAlgorithmsFlowchart100

7、1011001100010101001001001100101111101............1001011001100010101001001001110101111001............SelectionCrossoverMutationCurrentgenerationNextgenerationElitism7/24/202112GeneticAlgorithmsExample:Findthemax.ofthe“peaks”functionz=f(x,y)=3*(1-x)^2*exp(-(x

8、^2)-(y+1)^2)-10*(x/5-x^3-y^5)*exp(-x^2-y^2)-1/3*exp(-(x+1)^2-y^2).7/24/202113GeneticAlgorithmsDerivativesofthe“peaks”functiondz/dx=-6*(1-x)*exp(-x^2-(y+1)^2)-6*(1-x)^2*x*exp(-x^2-(y+1)^2)-10*(1/