After a large "teacher" neural network has been trained on labeled data, the
probabilities that the teacher assigns to incorrect classes reveal a lot of
information about the way in which the teacher generalizes. By training a small
"student" model to match these probabilities, it is p
该研究提出了一种名为 “Teaching what you Should Teach” 的策略，将其引入到知识蒸馏框架中。通过神经网络数据增强模块，该数据驱动蒸馏方法可以在查找适当的数据样本时帮助学生弥补其弱点，让学生成为全面优秀的人。此方法在目标识别、检测和分割任务上，通过详细的比较实验和探索性研究，证明它具有先进的性能和一定的推广能力。