Scalable Graph Hashing with Feature Transformation Model and Learning Feature Transformation oVx,define P(x)and Q(x): P(x)= 2(e2-1) -x2 e2+1-lx2 ;1川 ep e 2(e2-1 -x2 Q0x)= e2+1-x e p x;V -e e ;-1 ep oXi,Xj∈X P(x)Q)x -le 2e 2e 1x喔-l区房+2x ≈2e1 1 =2e喔 -1=5 日卡*·2元至Q0 Li (http://cs.nju.edu.cn/lvj) Learning to Hash LAMDA,CS.NJU 24/43Scalable Graph Hashing with Feature Transformation Model and Learning Feature Transformation ∀x, define P(x) and Q(x): P(x) = [s 2(e 2 − 1) eρ e −||x||2 F ρ x; r e 2 + 1 e e − ||x||2 F ρ ; 1] Q(x) = [s 2(e 2 − 1) eρ e −||x||2 F ρ x; r e 2 + 1 e e − ||x||2 F ρ ; −1] ∀xi , xj ∈ X P(xi) TQ(xj ) = 2[e 2 − 1 2e × 2x T i xj ρ + e 2 + 1 2e ]e − ||xi ||2 F +||xj ||2 F ρ − 1 ≈ 2e − ||xi ||2 F −||xj ||2 F +2x T i xj ρ − 1 = 2e − ||xi−xj ||2 F ρ − 1 = Seij Li (http://cs.nju.edu.cn/lwj) Learning to Hash LAMDA, CS, NJU 24 / 43