Synthesis of Data-Parallel Algorithms for Programmable Logic Devices

摘要

Most of the classifiers suffer from curse of dimensionality during classification of high dimensional image data. In this paper, we introduce a new supervised nonlinear dimensionality reduction (S-NLDR) algorithm called evolutionary strategy based supervised dimensionality reduction (ESSDR). The ESSDR method uses population based evolutionary strategy (ES) algorithm to find low dimensional embedded values of labeled data. Simulation studies on some well-known benchmark image data sets demonstrate that ESSDR produces better results in dimensionality reduction of labeled data as compare to other famous S-NDLR methods such as Weighted so, supervised locally linear embedding (SLLE), enhanced supervised locally linear embedding (ESLLE) and supervised local tangent space alignment (SLTSA).