The process of controlling how much of a model's memory capacity is available at each step of processing.