Flexible Auto-Weighted Local-Coordinate Concept Factorization: A Robust Framework for Unsupervised Clustering

Concept Factorization (CF) and its variants may produce inaccurate representation clustering results due to the sensitivity noise, hard constraint on reconstruction error, pre-obtained approximate similarities. To improve ability, a novel unsupervised Robust Flexible Auto-weighted Local-coordinate (RFA-LCF) framework is proposed for high-dimensional data. Specifically, RFA-LCF integrates robust flexible CF by clean data space recovery, sparse local-coordinate coding, adaptive weighting into unified model. improves representations enhancing robustness of noise errors, providing error optimizing locality jointly. For learning, clearly learns projection recover underlying space, then performed in projected feature space. also uses L2,1-norm based residue encode mismatch between recovered reconstruction, coding represent using few nearby basis concepts. auto-weighting, jointly preserves manifold structures concept new coordinate an manner minimizing errors data, anchor points coordinates. By updating preserving concepts coordinates alternately, abilities can be potentially improved. Extensive public databases show that delivers state-of-the-art compared with other related methods.