Assigning different amounts of parameters or computation to different layers rather than distributing them equally.