TY - JOUR
T1 - Fast Semi-Supervised Learning with Optimal Bipartite Graph
AU - He, Fang
AU - Nie, Feiping
AU - Wang, Rong
AU - Hu, Haojie
AU - Jia, Weimin
AU - Li, Xuelong
N1 - Publisher Copyright:
© 1989-2012 IEEE.
PY - 2021/9/1
Y1 - 2021/9/1
N2 - Recently, with the explosive increase in Internet data, the traditional Graph-based Semi-Supervised Learning (GSSL) model is not suitable to deal with large scale data as the high computation complexity. Besides, GSSL models perform classification on a fixed input data graph. The quality of initialized graph has a great effect on the classification result. To solve this problem, in this paper, we propose a novel approach, named optimal bipartite graph-based SSL (OBGSSL). Instead of fixing the input data graph, we learn a new bipartite graph to make the result more robust. Based on the learned bipartite graph, the labels of the original data and anchors can be calculated simultaneously, which solves co-classification problem in SSL. Then, we use the label of anchor to handle out-of-sample problem, which preserves well classification performance and saves much time. The computational complexity of OBGSSL is O(ndmt+nm^2)O(ndmt+nm2), which is a significant improvement compared with traditional GSSL methods that need O(n^2d+n^3)O(n2d+n3), where nn, dd, mm and tt are the number of samples, features anchors and iterations, respectively. Experimental results demonstrate the effectiveness and efficiency of our OBGSSL model.
AB - Recently, with the explosive increase in Internet data, the traditional Graph-based Semi-Supervised Learning (GSSL) model is not suitable to deal with large scale data as the high computation complexity. Besides, GSSL models perform classification on a fixed input data graph. The quality of initialized graph has a great effect on the classification result. To solve this problem, in this paper, we propose a novel approach, named optimal bipartite graph-based SSL (OBGSSL). Instead of fixing the input data graph, we learn a new bipartite graph to make the result more robust. Based on the learned bipartite graph, the labels of the original data and anchors can be calculated simultaneously, which solves co-classification problem in SSL. Then, we use the label of anchor to handle out-of-sample problem, which preserves well classification performance and saves much time. The computational complexity of OBGSSL is O(ndmt+nm^2)O(ndmt+nm2), which is a significant improvement compared with traditional GSSL methods that need O(n^2d+n^3)O(n2d+n3), where nn, dd, mm and tt are the number of samples, features anchors and iterations, respectively. Experimental results demonstrate the effectiveness and efficiency of our OBGSSL model.
KW - co-classification
KW - large scale data
KW - optimal bipartite graph
KW - out-of-sample
KW - Semi-supervised learning
UR - http://www.scopus.com/inward/record.url?scp=85112735946&partnerID=8YFLogxK
U2 - 10.1109/TKDE.2020.2968523
DO - 10.1109/TKDE.2020.2968523
M3 - 文章
AN - SCOPUS:85112735946
SN - 1041-4347
VL - 33
SP - 3245
EP - 3257
JO - IEEE Transactions on Knowledge and Data Engineering
JF - IEEE Transactions on Knowledge and Data Engineering
IS - 9
M1 - 8964288
ER -