Fast Discriminative Stochastic Neighbor Embedding Analysis

Feature is important for many applications in biomedical signal analysis and living system analysis. A fast discriminative stochastic neighbor embedding analysis (FDSNE) method for feature extraction is proposed in this paper by improving the existing DSNE method. The proposed algorithm adopts an alternative probability distribution model constructed based on its K-nearest neighbors from the interclass and intraclass samples. Furthermore, FDSNE is extended to nonlinear scenarios using the kernel trick and then kernel-based methods, that is, KFDSNE1 and KFDSNE2. FDSNE, KFDSNE1, and KFDSNE2 are evaluated in three aspects: visualization, recognition, and elapsed time. Experimental results on several datasets show that, compared with DSNE and MSNP, the proposed algorithm not only significantly enhances the computational efficiency but also obtains higher classification accuracy.