Scalable Graph Hashing with Feature Transformation Model and Leamning Sequential Learning Strategy o Then we obtain a generalized eigenvalue problem: K(X)RK(X)wt=XK(X)K(X)wt Define At =K(X)TRK(X)E Rmxm,then: A:=At-1-K(X)sgn(K(X)wt-1)sgn(K(X)wt-1)K(X) oKey component: A1=cK(X)SK(X) =c[K(X)P(X)Q(X)K(X)] s 日卡*·2元至风0 Li (http://cs.nju.edu.cn/lvj) Learning to Hash LAMDA,CS.NJU 27/43Scalable Graph Hashing with Feature Transformation Model and Learning Sequential Learning Strategy Then we obtain a generalized eigenvalue problem: K(X) T RtK(X)wt = λK(X) T K(X)wt Define At = K(X) T RtK(X) ∈ R m×m, then: At = At−1 − K(X) T sgn(K(X)wt−1)sgn(K(X)wt−1) T K(X) Key component: A1 = cK(X) TSeK(X) = c[K(X) T P(X) T ][Q(X) | {z } Se K(X)] Li (http://cs.nju.edu.cn/lwj) Learning to Hash LAMDA, CS, NJU 27 / 43