Semi-supervised learning via sparse model
Introduction
Semi-supervised learning (SSL) uses a large number of unlabeled samples together with a small number of labeled samples to build a better learner. It has attracted great research interest in both theory and practice in the past decade [1]. There exist various approaches in SSL to exploit unlabeled samples, showing incremental performance improvements [2]. Among them, graph-based methods, also known as manifold methods [3], [4], are widely used due to their good performances [3], [5], [6].
Through establishing the graph structure to represent adjacency relations among samples, the graph-based SSL methods can approximate the intrinsically low-dimensional manifold structure well [7]. Existing ways for graph construction include K-nearest-neighbor(KNN) method and ε-ball based method [5], [6]. In general, they construct the graph structure in two steps: (1) choosing adjacency samples needed to be connected, and (2) determining the edge weights [6]. graph, recently proposed by [4], is constructed at one run by solving each sample's sparse representation based on the rest of the entire sample set. It can surpass KNN method and ε-ball based method due to its characteristics of the high discriminative power, sparsity, and adaptive neighborhood [8]. However, these graph-based SSL methods have common drawbacks: high computational cost always comes together with the graph construction. When the adjacency relations among samples are computed to construct the graph structure, the complexity is [3], where n is the number of data samples and m is the dimensionality of the sample feature.
The graph-based SSL methods do not meet the cluster assumption. The cluster assumption states that if points are located in the same cluster, they are likely to belong to the same class. The success of SSL relies highly on certain semi-supervised assumptions (SSA) on the samples' distribution, which includes manifold and cluster assumptions [1]. The manifold assumption indicates that the high dimensional data lie on a low-dimensional manifold. Graph-based methods have not explored the cluster information of input distribution, so they may connect data points of different classes in cluster boundaries by the Euclidean distances of their feature vectors.
In order to overcome the above shortcomings of the graph-based SSL methods, we propose a new Semi-supervised learning method based on SPArse model, named SSPA. With the SSPA model, following advantages are achieved: Firstly adjacency relations among samples can be achieved by sparse matrix multiplication with a complexity of [9], where s is the average number of nonzero elements in each sparse representation and far less than the feature dimensionality m, so . Secondly, we explore the fact that the dictionary in SSPA model has the same cluster information as the training data samples, and thus we propose a structured dictionary learning algorithm to learn a structured dictionary which has explicit cluster information. With the structured dictionary the structured sparse model can be used in SSPA to gain better accuracy.
The main contribution of the work is that we propose a new SSPA model, with two kinds of dictionaries: non-structured [10] and structured. The SSPA model has a lower computational complexity than the graph-based SSL methods. With the structured dictionary the SSPA model can utilize the cluster information to improve the classification accuracy. The experimental results demonstrate that the proposed methods are efficient and outperform state-of-the-art graph-based SSL methods.
Section snippets
Related work
This work is closely related to the graph-based SSL methods and sparse model.
SSPA model
This section elaborates on the formulation of the SSPA model, and theoretically proves that the proposed SSPA model is capable of depicting the cluster and manifold structures of data samples.
Experiments
In this section, we conduct experiments to evaluate the accuracy and efficiency of the SSPA model at first, and then study the parameter sensitivity.
Conclusion and future work
In this paper, we propose a new model, SSPA, for semi-supervised learning based on sparse model. Since we compute adjacency matrix by a sparse matrix multiplication, the proposed method has lower computational cost than graph-based SSL methods. In order to meet the cluster assumption we implement SSPA algorithms with structured dictionary, besides non-structured one. With cluster information adjacency links among different clusters can be removed, and that can improve the accuracy of SSL.
Acknowledgments
This work was supported in part by the National Nature Science Foundation of China (61173054,61273247), and the National Key Technology Research and Development Program of China (2012BAH39B02).
Yu Wang received her Master's degree from Harbin Institute of Technology, China, in 2009. She is a Ph.D. candidate in the Institute of Computing Technology, Chinese Academy of Sciences (ICT-CAS), China.
Her current research interests include pattern recognition and content-based multimedia retrieval and indexing.
References (34)
- et al.
Sparse two dimensional local discriminant projections for feature extraction
Neurocomputing
(2011) - et al.
Semi-supervised learning via regularized boosting working on multiple semi-supervised assumptions
IEEE Trans. Pattern Anal. Mach. Intell.
(2011) - C. Leistner, A. Saffari, J. Santner, H. Bischof, Semi-supervised random forests, in: Proceedings of the 12th...
- G.S. Mann, A. McCallum, Simple, robust, scalable semi-supervised learning via expectation regularization, in:...
- et al.
Learning with l1-graph for image analysis
IEEE Trans. Image Proc.
(2010) - X. Zhu, Z. Ghahramani, J. Lafferty, Semi-supervised learning using gaussian fields and harmonic functions, in:...
- et al.
Manifold regularizationa geometric framework for learning from labeled and unlabeled examples
J. Mach. Learn. Res.
(2006) - et al.
A global geometric framework for nonlinear dimensionality reduction
Science
(2000) - R. He, W.-S. Zheng, B.-G. Hu, X.-W. Kong, Nonnegative sparse coding for discriminative semi-supervised learning, in:...
- S. Williams, L. Oliker, R. Vuduc, J. Shalf, K. Yelick, J. Demmel, Optimization of sparse matrix-vector multiplication...
Learning a propagable graph for semisupervised learningclassification and regression
IEEE Trans. Knowl. Data Eng.
Supervised dictionary learning
Adv. Neural Inf. Proc. Syst.
Sparse approximation to the eigensubspace for discrimination
IEEE Trans. Neural Netw. Learn. Syst.
Cited by (5)
Graph-based semi-supervised learning: A review
2020, NeurocomputingExpert list-wise ranking method based on sparse learning
2016, NeurocomputingSemi-supervised image segmentation with globalized probability of boundary and simple linear iterative clustering
2018, ICNC-FSKD 2017 - 13th International Conference on Natural Computation, Fuzzy Systems and Knowledge DiscoveryAn image segmentation method based on Simple Linear Iterative Clustering and graph-based semi-supervised learning
2016, Proceedings of 2015 International Conference on Orange Technologies, ICOT 2015Semi-supervised two-dimensional manifold learning based on pair-wise constraints
2014, Proceedings of the 33rd Chinese Control Conference, CCC 2014
Yu Wang received her Master's degree from Harbin Institute of Technology, China, in 2009. She is a Ph.D. candidate in the Institute of Computing Technology, Chinese Academy of Sciences (ICT-CAS), China.
Her current research interests include pattern recognition and content-based multimedia retrieval and indexing.
Sheng Tang received his Ph.D. degree in computer application technology at the Institute of Computing Technology, Chinese Academy of Sciences (ICT-CAS), China, in 2006. He is now an associate professor in the ICT-CAS. He visited National University of Singapore (NUS) for participating in TREC Video Retrieval Evaluation (TRECVid) tasks, in 2006. From 2006 to 2008, he was the TRECVid team (MCG-ICT-CAS) leader of the Multimedia Computing Groupat ICT-CAS. From February 2009 to February 2010, he worked as a visiting research fellow in NUS under the instruction of Prof. Chua Tat-Seng. His current research interests are in the fields of pattern recognition and machine learning, multimedia information processing, in particular, on indexing, retrieval and extraction of information in image and video.
Dr. Tang is a member of ACM, IEEE, and senior member of China Computer Federation (CCF). He serves as the reviewer for a number of famous international conferences and journals, such as IEEE Transactions on Information Forensics & Security, Neurocomputing, Elsevier Data & Knowledge Engineering (DKE), Journal of Visual Communication and Image Representation, Multimedia Tools and Applications, Journal of Computer Science and Technology, Chinese Journal of Computers, Chinese Journal of Computer Research and Development, etc. His homepage: http://mcg.ict.ac.cn/people/shengtang.htm.
Yan-Tao Zheng received his Ph.D. from National University of Singapore (with Best Ph.D. Thesis Award) and B.E. (with 1st class Hons) from Nanyang Technological University, Singapore. He is a research scientist at Institute for Infocomm Research, Singapore. His research interests include geo-mining in multimedia, image annotation and video search. During his attachment at Google Inc., in 2008, he developed a world-scale landmark recognition engine together with Google engineers, which has been highly praised and well publicized.
Dr. Zheng is the recipient of a number of international awards, including Champion of Star Challenge, Microsoft Research Fellowship, IBM Waston Emerging Multimedia Leaders, and so on. He has served as program committee member and reviewer of a number of prestigious international conferences and journals.
Yong-Dong Zhang received his Ph.D. degree in Electronic Engineering from Tianjin University, Tianjin, China, in 2002. Dr. Zhang is currently a Professor in the ICT-CAS, China. His research interests are in the field of video coding and transcoding, video analysis and retrieval, and universal media access.
Jin-Tao Li received his Ph.D. degree in computer science at the Institute of Computing Technology, Chinese Academy of Sciences (ICT-CAS), China, in 1989. He spent the years of 1989, 1990 at Academy of Sciences of the Czech Republic as a visiting scholar.
Dr. Li is currently a professor and Ph.D. supervisor in the ICT-CAS, China. His current research interests focus on the areas of digital image processing, video coding, multimedia content analysis and retrieval, digital watermarking, and network computing technology, etc.