Conditional Generative Data-free Knowledge Distillation

Xinyi Yu, Ling Yan, Yang Yang +2

Knowledge distillation has made remarkable achievements in model compression. However, most existing methods require the original training data, which is usually unavailable due to…