Correntropy based nonnegative matrix factorization : algorithms and clustering applications

In this thesis, to improve existing correntropy based nonnegative matrix factorization (NMF) algorithms and develop new methods for enlarging the range and enhancing the performance in clustering tasks, three novel correntropy based NMF algorithms are proposed, which are respectively the correntropy based graph regularized concept factorization algorithm, the correntropy based orthogonal nonnegative matrix tri-factorization algorithm, and the correntropy based semi-supervised nonnegative matrix factorization algorithm. The half-quadratic optimization technique is adopted to solve the optimization problems of the proposed algorithms, and the multiplicative update rules are derived. The new algorithms are analyzed from different aspects such as convergence, robustness, computational complexity, and relationships with several previous NMF methods. Experimental results demonstrate the effectiveness and robustness of the proposed algorithms for clustering applications on real world datasets compared with several state-of-the-art methods.Doctor of Philosoph