PeerJ Computer Science, Год журнала: 2025, Номер 11, С. e2756 - e2756
Опубликована: Март 24, 2025
Neural networks are a state-of-the-art approach that performs well for many tasks. The activation function (AF) is an important hyperparameter creates output against the coming inputs to neural network model. AF significantly affects training and performance of Therefore, selecting most optimal processing input data in important. Determining often difficult task. To overcome this difficulty, studies on trainable AFs have been carried out literature recent years. This study presents different apart from fixed or approaches. For purpose, cyclically switchable convolutional (AFCS-CNN) model structure proposed. AFCS-CNN does not use value during training. It designed self-regulating by switching proposed based logic starting with selection among next depending decrease Any (CNN) can be easily used structure. In way, simple but effective perspective has presented. study, first, ablation using Cifar-10 dataset determine CNN models specific hyperparameters After were determined, expansion experiments datasets results showed achieved success datasets.
Язык: Английский