ABSTRACT This systematic review investigates the use of quantum activation functions in neural networks applied to computational science models. The results indicate that this field is currently in a transitional phase, where hybrid classical–quantum models predominate due to hardware and implementation limitations. Adapting classical activation functions to quantum equivalents emerges as a crucial step toward building effective hybrid networks, with some studies proposing promising approaches. At the same time, the initial explorations of fully quantum models demonstrate potential for future applications, although they largely remain at a conceptual or experimental stage. This work highlights the gradual transition from classical to quantum computing in machine learning, emphasizing both the current challenges and the opportunities for future research.
Santos et al. (Mon,) studied this question.