Two-Sided Sparse Learning with Augmented Lagrangian Method

Xiaohua Xu, Baichuan Fan, Ping He, Yali Liang, Yuan Lou, Zhijun Zhang, Xincheng Chang

Abstract: In this paper, we propose a novel sparse learning model, named Two-Sided Sparse Learning with Augmented Lagrangian Method, and apply it to the classification problem. Existing dictionary learning method only emphasizes the sparsity of cases, but neglect the sparsity of features. In the context of classification, it is crucial to take into account the correlation among features and find the most representative features in a class. By representing training data as sparse linear combination of rows and columns in dictionary, this model can be more suitable for classification problem. Experimental results demonstrate that our model achieves superior performance than the state-of-the-art classification methods on real-world datasets.