摘要
Aneuralnetwork(NN)isapowerfultoolforapproximatingboundedcontinuousfunctionsinmachinelearning.TheNNprovidesaframeworkfornumericallysolvingordinarydifferentialequations(ODEs)andpartialdifferentialequations(PDEs)combinedwiththeautomaticdifferentiation(AD)technique.Inthiswork,weexploretheuseofNNforthefunctionapproximationandproposeauniversalsolverforODEsandPDEs.ThesolveristestedforinitialvalueproblemsandboundaryvalueproblemsofODEs,andtheresultsexhibithighaccuracyfornotonlytheunknownfunctionsbutalsotheirderivatives.ThesamestrategycanbeusedtoconstructaPDEsolverbasedoncollocationpointsinsteadofamesh,whichistestedwiththeBurgersequationandtheheatequation(i.e.,theLaplaceequation).
出版日期
2019年02月12日(中国期刊网平台首次上网日期,不代表论文的发表时间)