Scalable Graph Hashing with Feature Transformation Model and Leamning Sequential Learning Strategy Adopting the residual matrix: Rt=S-∑ sgn(K(X)w:)sgn(K(X)wi) =1,i≠t we can learn all the W=[wil1 for multiple rounds. o This can further improve the accuracy. o We continue it for one more round to get a good tradeoff between accuracy and speed. 日卡三4元，互)Q0 Li (http://cs.nju.edu.cn/lvj) Learning to Hash LAMDA,CS.NJU 28/43Scalable Graph Hashing with Feature Transformation Model and Learning Sequential Learning Strategy Adopting the residual matrix: Rt = cSe − Xc i=1,i6=t sgn(K(X)wi)sgn(K(X)wi) T we can learn all the W = {wi} c i=1 for multiple rounds. This can further improve the accuracy. We continue it for one more round to get a good tradeoff between accuracy and speed. Li (http://cs.nju.edu.cn/lwj) Learning to Hash LAMDA, CS, NJU 28 / 43