unsupervis featur select multi cluster data deng cai chiyuan zhang xiaofei data analysi task confront high dimension data featur select techniqu design find relev featur subset origin featur facilit cluster classif retriev paper featur select problem unsupervis learn scenario difficult due absenc class label guid search relev inform featur select problem essenti combinatori optim problem computation expens tradit unsupervis featur select method address issu select top rank featur base score comput independ featur approach neglect correl featur produc optim featur subset inspir recent develop manifold learn regular model subset select propos paper approach call multi cluster featur select mcf unsupervis featur select specif select featur multi cluster structur data preserv optim problem effici solv involv spars eigen problem regular squar problem extens experiment result real life data set demonstr superior propos algorithm 