t-SNE Based on Sobol Sequence Initialized Exchange Market Algorithm

T-distributed stochastic neighbor embedding (t-SNE) is a nonlinear dimensionality reduction method in manifold learning in the field of artificial intelligence (AI). It is very suitable for the visualization of high-dimensional data. It has now been widely used in image processing, natural language processing, speech processing, and gene data analysis. However, t-SNE also has some problems, for example, the solution process depends on gradient information, which is prone to fall into local optimum. At present, many scholars have studied t-SNE and have tried to employ t-SNE to solve specific problems of different industries, but most of them have not done related work on the optimization of its solution process. In this paper, a Sobol sequence initialized exchange market algorithm (SSEMA) has been proposed to replace the gradient dependent solution method in the original t-SNE, which can find a more reliable global optimal solution by comparing with just considering about gradient information.