Softmax Regression Based on Bacterial Foraging Optimization Algorithm with t-Distribution Parameters

Abstract

Softmax regression is a supervised multi-class nonlinear classification algorithm in machine learning. Sometimes it is also used in regression problems. It is mainly used in transfer learning, knowledge distillation and meta learning in artificial intelligence. Softmax optimizer is gradient descent method or random gradient descent method, but its loss function is a multi peak, strong nonlinear function, and the optimal solution is only the local optimal solution rather than the global optimal solution, and depends on the gradient. So we use a swarm intelligence optimization algorithm which is independent of gradient and can find the global optimal solution to optimize. This article will introduce the bacterial foraging optimization algorithm.