Adaptive Semi-supervised Learning Based On Low Rank Representation

Data analysis has been a basic and important research topic in the fields of pattern recognition and machine learning. With the rapid development of computer technology, it is easy to get a large number of unlabeled data, but it takes a lot of manpower and time to label the data. So how to learn useful knowledge from a small amount of labeled data and a large number of unlabeled data is the key and difficult point in the fields of data analysis,Semi-supervised Learning is an effective solution. Because of its good classification performance, simple calculation and other advantages, graph-based method has been a mainstream in Semi-supervised Learning. This paper does some research on graph-based Semi-supervised Learning, the main contents of this article are as follows:1. The traditional Semi-supervised Learning needs to set parameter manually, which leads to poor adaptiveness. This paper makes a research on Non-negative Low Rank and Sparse graph for Adaptive Semi-supervised Learning(NLRS-ASL). The algorithm mainly adds the sparse constraints on low rank representation, it simultaneously derives the graph structure and the graph weights by solving a problem of non-negative low rank and sparse representation model, which can avoid the intervention of manual parameter, In addition,the algorithm takes the global and local structure of the data into the consideration at the same time, then uses the graph preserving principle to constraint that the similar samples should have similar category information, which ensures better discrimination.Experimental results on the ORL, Extended Yale B and PIE databases demonstrate that the proposed algorithm is of high classification accuracy and strong robustness.2. When there is a small amount of labeled training samples, the Sparse Representation-based Fisher Discriminant Criterion algorithm(SRC-FDC) may lead to a significant performance degradation. In order to solve this problem, this paper makes a research on the Sparse Representation-based Semi-supervised Discriminant Analysis algorithm(SRC-SDA). Compared to SRC-FDC algorithm, SRC-SDA utilizes the idea of Semi-supervised Learning, even if there is a small amount of labeled training samples,SRC-SDA can combine the unlabeled samples' information to find a more discriminant projection matrix, which ensures the algorithm is robust. Experimental results on the COIL,Extended Yale B and PIE databases demonstrate that the proposed algorithm is of strong feasibility and effectiveness.