Locally Linear Embedding Based on Neiderreit Sequence Initialized Pelican Optimization Algorithm

Abstract

Locally linear embedding (LLE) algorithm is an important nonlinear dimensionality reduction method in the field of manifold learning of artificial intelligence (AI). The core idea of LLE is to find a homeomorphic mapping that can ensure that the linear algebraic structure of low-dimensional spatial data projected from high-dimensional spatial data remains unchanged. LLE contains two optimization sub-problems, and gradient descent (GD) or stochastic gradient descent (SGD) method is often used for parameter optimization. However, gradient dependent method is easy to fall into the trap of local minimum, resulting in failure to find the global optimal solution. In order to overcome the above problems, in this paper, a new Neiderreit sequence initialized pelican optimization algorithm (NSPOA) has been proposed, to help LLE find the global optimal parameter properly. Numerical experiment has shown that NSPOA has higher operation efficiency than traditional gradient dependent optimization methods.