摘要
Inapplicationsoflearningfromexamplestoreal-worldtasks,featuresubsetselectionisimportanttospeeduptrainingandtoimprovegeneralizationperformance.Ideally,aninductivealgorithmshouldusesubsetoffeaturesassmallaspossible.Inthispaperhowever,theauthorsshowthattheproblemofselectingtheminimumsubsetoffeaturesisNP-hard.Thepaperthenpresentsagreedyalgorithmforreaturesubsetselection.Theresultofrunningthegreedyalgorithmonhand-writtennumeralrecognitionproblemisalsogiven.
出版日期
1997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)