Kernel locality preserving projections algorithm based on T-class weighted kernel functions with center kernel alignment
Zhu Shiming, Wang Zhengqun, Xia Jin · 2023
The current data generally has a high dimension and contains many features unrelated to category labels.On the basis of local preserving projection,we proposed a kernel local preserving projection algorithm based on T-class weighted kernel function of kernel alignment.Firstly,CKA algorithm was used to calculate the weight of each kernel matrix,which was used as linear combination coefficient.Combined kernel function was constructed by combining different types of kernel functions to improve the recognition rate and anti-noise ability.Secondly,we use the t distribution in probability statistics to construct the T-type kernel function,which further improves the combined kernel function.Finally,we use the kernel local preservation projectin for identification,and verify it on several data sets.The experimental results prove the effectiveness of the proposed method.