Improved Salp Swarm Algorithm with Space Transformation Search for Training Neural Network

Abstract

Swarm-based algorithm is best suitable when it can perform smooth balance between the exploration and exploitation as well as faster convergence by successfully avoiding local optima entrapment. At recent time, salp swarm algorithm (SSA) is developed as a nature-inspired swarm-based algorithm. It can solve continuous, nonlinear and complex in nature day-to-day life optimization problems. Like many other optimization algorithms, SSA suffers with the problem of local stagnation. This paper introduces an improved version of the SSA, which improves the performance of the existing SSA by using space transformation search (STS). The proposed algorithm is termed as STS-SSA. The STS-SSA enhances the exploration and exploitation capability in the search space and successfully avoids local optima entrapment. The STS-SSA is evaluated by considering the IEEE CEC 2017 standard benchmark function set. The efficiency and robustness of the proposed STS-SSA are measured using performance metrics, convergence analysis and statistical significance. A demonstration is given as an application of the proposed algorithm for solving a real-life problem. For this purpose, the multi-layer feed-forward network is trained using the proposed STS-SSA. The experimental results demonstrate that the developed STS-SSA can be used for solving optimization problems effectively.