Sigmoidal learning rate optimizer for deep neural network training using a two-phase adaptation approach

David Macêdo, Cleber Zanchettin, Teresa Bernarda Ludermir

Abstract: Highlights•TASO is a novel, two-phase automated learning rate scheduler.•The method outperforms traditional adaptive learning rate methods on image and text datasets.•TASO reduces the need for manual hyperparameter tuning, improving overall effectiveness.