Abstract

The performance of complex recurrent networks heavily relies on the selection of an appropriate activation function. Some established activation functions, known to enhance complex-valued recurrent networks, are chosen based on specific network properties. Exploring alternative approaches to complex activation holds promise for improving the performance of complex recurrent networks. This study introduces a novel activation function that leverages the structural properties of the Hopf bifurcation. The proposed activation function is evaluated using the Mackey Glass and Copy Memory datasets, commonly employed for analyzing recurrent networks. Two experiments are conducted: one to determine an appropriate configuration for the activation function and another to compare it against five established activations. The results indicate that the Hopf bifurcation shows promise as a complex activation, delivering comparable or better performance compared to more established functions.