Exponential Graph Regularized Non-Negative Low-Rank Factorization for Robust Latent Representation

Non-negative matrix factorization (NMF) is a fundamental theory that has received much attention and widely used in image engineering, pattern recognition other fields. However, the classical NMF limitations such as only focusing on local information, sensitivity to noise small sample size (SSS) problems. Therefore, how develop improve performance robustness of algorithm worthy challenge. Based bottlenecks above, we propose an exponential graph regularization non-negative low-rank (EGNLRF) combining sparseness, low rank exponential. Firstly, based assumption data corroded, decompose given raw item with error fitting matrix, applying constraint denoising data. Then, perform resulting from which derive low-dimensional representation original matrix. Finally, use for embedding maintain geometry between samples. The terms are exponentiated cope SSS problems nearest neighbor sensitivity. above three steps will be incorporated into joint framework validate optimize each other; therefore, can learn latent representations undisturbed by preserve structure known We conducted simulation experiments different datasets verified effectiveness comparing proposed lasting ones related NMF, embedding.