semi-supervisedlearningwithkerneldensityestimation

Insufficiencyoflabeledtrainingdataisamajorobstacleforautomaticallyannotatinglarge-scalevideodatabaseswithsemanticconcepts.Existingsemi-supervisedlearningalgorithmsbasedonparametricmodelstrytotacklethisissuebyincorporatingtheinformationinalargeamountofunlabeleddata.However,theyarebasedon