Cam-weighted local tangent space alignment for dimension reduction

Manifold learning works well in discovering the intrinsic structures embedded in the high-dimensional coordinates and most of the methods lie on the distance metric. This article proposes an unsupervised learning algorithm called cam-weighted distance local tangent space alignment (Cam-WLTSA) which performs well with artificial sample data sets. The cam-weighted distance, first using in the nearest neighbour classification, plays a significant role in this algorithm, which avoiding inappropriate neighbour searching. This algorithm can discover the intrinsic structures which can be used for classification and recognition. Simulation studies demonstrate Cam-WLTSA can give better results in dimensional reduction than LTSA with some artificial sample data sets.