t-SNE Based on Hammersley Sequence Initialized Genghis Khan Shark Optimizer

Manifold learning is a nonlinear dimensionality reduction technique that preserves the original data features and obtains the intrinsic probability features and distribution structure of high-dimensional data through dimensionality reduction and simplification processing. t-distribution stochastic neighbor embedding (t-SNE) is an important algorithm in manifold learning that can visualize high-dimensional data by preserving the local probability distribution structure of the data in a low-dimensional space. However, the loss function of t-SNE is a strongly nonlinear non-convex function, and parameter optimization often faces difficulties in global optimization and high computational complexity, which limits the efficiency and scalability of the algorithm. To address these issues, in this paper, a derivative-free global optimization algorithm as well as a swarm intelligence method, namely Hammersley sequence initialized Genghis Khan shark optimizer (HSGKSO) has been proposed. The proposed algorithm can achieve stronger global search capability and faster convergence speed, and its effectiveness in finding the optimal solution of t-SNE loss function has been verified through numerical experiments.