This type оf regulаrizаtiоn invоlves the rаndom removal of neurons at each step of gradient descent, tracking the fraction of the time they are present, and adjusting the weights based on the amount of time they were present during training.