Stochastic Neighbor Embedding Based on Faure Sequence Initialized Chimp Optimization Algorithm

Abstract

Stochastic neighbor embedding (SNE) is a nonlinear dimensionality reduction method of classical manifold learning in the field of artificial intelligence (AI). The main idea of SNE is to pursue the homeomorphism and isomorphism of the probability space for data points in both high-dimensional space and low-dimensional space. In SNE, Kullback-Leibler (KL) divergence is usually used to measure the approximation between two probability distributions, and it is usually solved by gradient descent method (GD) or stochastic gradient descent method (SGD). However, gradient dependent methods has the following shortcomings: easy to fall into the local minimum trap; the closer to the optimal value, the more possibility to appear sawtooth effect. In order to overcome the above shortcomings, in this paper, a Faure sequence initialized chimp optimization algorithm (FSChOA) has been proposed, which belongs to swarm intelligence algorithm and is also a derivative free optimization algorithm. By experimental comparison, it can be concluded that FSChOA is has more advantages in solving KL divergence for SNE.