How a model's decision-making strategy changes over training iterations, affecting which samples it generates and with what probability.