Comparison and Combination of Activation Functions in Broad Learning System
Lili Xu, C. L. Philip Chen · 2020
Activation function is a crucial component in artificial neural networks for its capability of converting linear function of input to complex nonlinear expression. It also plays an important role generating enhancement nodes in broad learning system(BLS). In this paper, we perform the comparison of 20 popular activation functions on different datasets in classification and regression. Among all selected activation functions, sigmoid leads to faster training process and greater approximation capability than others in general tasks. Meanwhile, the statistical analysis demonstrates that the type of activation function does not affect the performance of BLS too much. Afterwards, we assemble some best-performing activation functions to form a combination within convex restriction, which achieves better performance than corresponding base activation functions in standard BLS.