Abstract.WedescribeseveralimprovementstoFreundandSchapire’sAdaBoostboostingalgorithm,particularlyinasettinginwhichhypothesesmayassignconfidencestoeachoftheirpredictions.WegiveasimplifiedanalysisofAdaBoostinthissetting,andweshowhowthisanalysiscanbeusedtofindimprovedparametersettingsa