Hidden Markovmodels(HMMs)arewidelyusedprobabilisticmodelsofsequentialdata.Aswithother
probabilistic models,theyrequirethespecification oflocalconditionalprobabilitydistributions,whose
assessment canbetoodifficult anderror-prone,especiallywhendataarescarceorcostlytoacquire.The
imprecise HMM(iHMM)generalizesHMMsbyallowingthequantification tobedonebysetsof,instead
of single,probabilitydistributions.iHMMshavetheabilitytosuspendjudgmentwhenthereisnot
enough statisticalevidence,andcanserveasasensitivityanalysistoolforstandardnon-stationary
HMMs. Inthispaper,weconsideriHMMsunderthestrongindependenceinterpretation,forwhichwe
develop efficient inferencealgorithmstoaddressstandardHMMusagesuchasthecomputationof
likelihoods andmostprobableexplanations,aswellasperforming filtering andpredictiveinference.
Experiments withrealdatashowthatiHMMsproducemorereliableinferenceswithoutcompromising
the computationalefficiency.
Arewidelyusedprobabilisticmodelsofsequentialdata Markovmodels (HMMs) ที่ซ่อนอยู่ Aswithotherรุ่น probabilistic, theyrequirethespecification oflocalconditionalprobabilitydistributions มีประเมิน canbetoodifficult anderror เสี่ยง especiallywhendataarescarceorcostlytoacquire ที่imprecise HMM (iHMM) generalizesHMMsbyallowingthequantification tobedonebysetsof แทนของ single,probabilitydistributions.iHMMshavetheabilitytosuspendjudgmentwhenthereisnotพอ statisticalevidence, andcanserveasasensitivityanalysistoolforstandardnon เครื่องเขียนHMMs. Inthispaper, weconsideriHMMsunderthestrongindependenceinterpretation, forwhichweพัฒนา inferencealgorithmstoaddressstandardHMMusagesuchasthecomputationof ที่มีประสิทธิภาพandmostprobableexplanations likelihoods, aswellasperforming กรอง andpredictiveinferenceWithrealdatashowthatiHMMsproducemorereliableinferenceswithoutcompromising ทดลองcomputationalefficiency
การแปล กรุณารอสักครู่..