A method used in neural network training aimed at managing computational complexity and improving model efficiency by adjusting the size of the model layers.
