Data Representation by Joint Hyper graph Embedding and Sparse Coding
Abstract : Matrix factorization (MF), a popular unsupervised learning technique for data representation, has been widely applied in data mining and machine learning.
According to different application scenarios, one can impose different constraintson the factorization to find the desired basis, which captures high-level semantics for the given data, and learns the compact representation corresponding to the basis.
In this work, we propose a novel MF framework called Joint Hyper graph Embedding and Sparse Coding (JHESC), in which the obtained basis captures high-order semantic information in data. Specifically, we first propose a new hyper graph learning model to obtain a more discriminative basis by hyper graph-based Laplacian Eigen map, then sparse coding is conducted on the learned basis such that the new representation has stronger identification capability.
In addition, we extend the proposed method to the reproducing kernel Hilbert space for dealing with nonlinear data more effectively. Extensive experimental results on data clustering demonstrate that the proposed method consistently outperforms the other state-of-the-art matrix factorization methods
We note that almost all previous work on MF in data mining has ignored to find such a basis, which can carry high-order semantics in the data.
? Semi supervised learning is an effective technique to represent the intrinsic features of a hyper spectral image (HSI),which can reduce the cost to obtain the labeled information of samples.
? However, traditional semi supervised learning methods fail to consider multiple properties of an HSI, which has restricted the discriminant performance of feature representation.
? In this article , we introduce the hyper graph into semi supervised learning to reveal the complex multi structures of an HSI, and construct a semi supervised discriminant hyper graph learning (SSDHL) method by designing an intraclass hypergraph and an interclass graph with the labeled samples. SSDHL constructs an unsupervised hypergraph with the unlabeled samples
? we propose a novel approach to construct a hyper graph from the perspective on probability, which can adaptively determine neighbors of the center point as well as give a weight for each data point indifferent hyper edges.
? Although the graph is an effective tool in modeling intrinsic structures of data, simply modeling pair wise relationships among the objects in the real world may lead to a loss of some information, which is crucial to our learning tasks.
? This is because in practical problems, objects tend to exhibit high-order relationships instead of pair wise ones
? we propose a semi supervised hyper graph learning model to extract the low-dimensional features of an HSI with both the labeled and unlabeled samples.
? Semi supervised learning was proposed to simultaneously utilize the labeled and unlabeled samples to construct the DR models
? In this article, we proposed a semi supervised discriminant hyper graph learning (SSDHL) method to obtain the effective low-dimensional features for HSI classification.
? A series of work showed good performance of NMF and its variants in various domains
? Better bandwidth tuning and using the proposed hyper graph construction would achieve better clustering performance for KJHESC.
? He proposed hyper graph learning model to construct a hyper graph, against the same compared methods performed on microarray data sets.
? We mark the best results in bold.
? It is clear that our methods obtain better performance on those datasets.
|