Incremental locality preserving nonnegative matrix factorization
Jianwei Zheng, Yu Chen, Yiting Jin, Wanliang Wang · 2013
Recently nonnegative matrix factorization (NMF) has become a popular dimension reduction method and it has been successfully applied to image processing and pattern recognition. In this paper, we propose an incremental locality preserving nonnegative matrix factorization (ILPNMF) method, which is aimed to discover the manifold structure embedded in high-dimensional space that deals well with large scale data. By assuming that the newly added samples do not change the encoding vectors of old samples, we present a cost function for online learning. Then we use projected gradient method to solve the update rule of the cost function. Experimental results show that ILPNMF provides a better parts-based representation compared with INMF and it is faster than the batch one LPNMF.