[1]XU Guangsheng,WANG Shitong.Incomplete multi-view subspace learning through dual low-rank decompositions[J].CAAI Transactions on Intelligent Systems,2022,17(6):1084-1092.[doi:10.11992/tis.202107002]
Copy
CAAI Transactions on Intelligent Systems[ISSN 1673-4785/CN 23-1538/TP] Volume:
17
Number of periods:
2022 6
Page number:
1084-1092
Column:
学术论文—机器学习
Public date:
2022-11-05
- Title:
-
Incomplete multi-view subspace learning through dual low-rank decompositions
- Author(s):
-
XU Guangsheng1; WANG Shitong2
-
1. School of Artificial Intelligence and Computer Science, Jiangnan University, Wuxi 214122, China;
2. Key Laboratory of Media Design and Software Technology of Jiangsu Province, Jiangnan University, Wuxi 214122, China
-
- Keywords:
-
subspace learning; supervised learning; incomplete multi-view; latent factors; low-rank constraint; dual low-rank decompositions; feature alignment; low-dimensional feature
- CLC:
-
TP181
- DOI:
-
10.11992/tis.202107002
- Abstract:
-
Multi-view data are very common in real-world applications. Different viewpoints and sensors tend to facilitate better data representation. However, data from various perspectives show a significant variation. Especially when only incomplete multi-view data are available, the corresponding multi-view learning may result in poor performance or even training failure. This study proposes a multi-view learning algorithm called IMSL (Incomplete Multi-View Subspace Learning through Dual Low-Rank Decompositions) to tackle this issue. The proposed algorithm addresses the incomplete multi-view problem in two ways: (1) Latent factors are introduced into a dual low-rank decomposition subspace framework to mine missing information in the multi-view data. (2) IMSL seeks a more robust subspace through pre-learned low-dimensional features of multi-view data. Furthermore, the supervised data are used to guide dual low-rank decompositions. Experimental results show that the proposed algorithm outperforms the previous multi-view subspace learning algorithms on the adopted incomplete multi-view datasets. Pre-learning low-dimensional features of multi-view data, on the other hand, can improve robustness, and dual low-rank decomposition can be guided in a supervised manner.