Improved Manifold Learning Algorithm for Data Dimension Reduction Based on KNN
Liming Liang, Falu Weng, Zhaoyang Chen, Zhen Zhong · 2014
In this paper, a new multi-manifold learning algorithm based on KNN algorithm is proposed in order to provide manifold learning model automatic parameters selection strategy.Basic ideas for such a algorithm is constructing a weighted norm as the variable of the intrinsic low dimensions expression function, and then optimizing the function's variables and getting a automatic selection of the size of the intrinsic low dimensions and the neighborhood in the manifold learning algorithm model.After a series of numerical experiments on simulated and experimental, results proves the feasibility and effectiveness of the algorithm.