scalabl similar search optim kernel hash junfeng wei liu shih fu chang scalabl similar search core larg scale learn data mine applic recent research result demonstr promis approach creat compact effici hash code preserv data similar effici refer low correl low redund gener code exist hash method design vector data paper develop hash algorithm creat effici code larg scale data gener format kernel function includ kernel vector graph sequenc set start idea analog spectral hash formul solut propos kernel base hash function explicitli repres optim directli appli comput compact hash code sampl gener format incorpor effici techniqu nystrom approxim reduc time space complex index search make algorithm scalabl huge data set import advantag method abil handl divers type similar actual task requir includ featur similar semant similar label consist evalu method vector vector data set larg scale million sampl comprehens result show propos method outperform state art approach task signific gain task 