Scalable Graph Hashing with Feature Transformation Model and Leamning Objective Function min∑y(Si-bb)2 0Sj=2S-1∈(-1， Hash function:h(xi)=sgn(Wj(xi,j)+bias) Vxi,map it into Hamming space as bi=[h1(xi),...,he(xi)] o Vx,define:K(x)= [p(x,x1)-1(x,x)/m,,(x,xm)-1(x,xm)/m Objective function with the parameter W E Rexm: min lles-sgn(K(x)WT)sgn(K (x)WT)T s.t.WK(X)TK(X)wT=I 日卡三4元，互Q0 Li (http://cs.nju.edu.cn/lwj) Learning to Hash LAMDA,CS.NJU 23/43Scalable Graph Hashing with Feature Transformation Model and Learning Objective Function min P ij (Seij − 1 c b T i bj ) 2 Seij = 2Sij − 1 ∈ (−1, 1]. Hash function: hk(xi) = sgn( Pm j=1 Wijφ(xi , xj ) + bias) ∀xi , map it into Hamming space as bi = [h1(xi), · · · , hc(xi)]T ∀x, define: K(x) = [φ(x, x1) − Pn i=1 φ(xi , x1)/n, . . . , φ(x, xm) − Pn i=1 φ(xi , xm)/n] Objective function with the parameter W ∈ R c×m: min W ||cSe − sgn(K(X)WT )sgn(K(X)WT ) T ||2 F s.t. WK(X) T K(X)WT = I Li (http://cs.nju.edu.cn/lwj) Learning to Hash LAMDA, CS, NJU 23 / 43