A Self-Organising Multi-Manifold Learning Algorithm

This paper presents a novel self-organising multi-manifold learning algorithm to extract multiple nonlinear manifolds from data. Extracting these sub-manifolds or manifold structure in the data can facilitate the analysis of large volume of data and discover their underlying patterns and generative causes. Many real data sets exhibit multiple sub-manifold structures due to multiple variations as well as multiple modalities. The proposed learning scheme can learn to establish the intrinsic manifold structure of the data. It can be used in either unsupervised or semi-supervised learning environment where ample unlabelled data can be effectively utilized. Experimental results on both synthetic and real-world data sets demonstrate its effectiveness, efficiency and promising potentials in many big data applications.