1 paper
Haoyu He, Xingjian Shi, Jonas Mueller +3
We aim to identify how different components in the KD pipeline affect the resulting performance and how much the optimal KD pipeline varies across different datasets/tasks, such as…