Explaining Knowledge Distillation by Quantifying the Knowledge
This provides insights into knowledge distillation mechanisms for researchers in model compression and interpretability, but it is incremental as it focuses on explaining existing methods rather than introducing new ones.
The paper tackles the problem of interpreting why knowledge distillation works by quantifying task-relevant and task-irrelevant visual concepts in deep neural networks, verifying three hypotheses that show distillation leads to learning more concepts, learning them simultaneously rather than sequentially, and providing more stable optimization.
This paper presents a method to interpret the success of knowledge distillation by quantifying and analyzing task-relevant and task-irrelevant visual concepts that are encoded in intermediate layers of a deep neural network (DNN). More specifically, three hypotheses are proposed as follows. 1. Knowledge distillation makes the DNN learn more visual concepts than learning from raw data. 2. Knowledge distillation ensures that the DNN is prone to learning various visual concepts simultaneously. Whereas, in the scenario of learning from raw data, the DNN learns visual concepts sequentially. 3. Knowledge distillation yields more stable optimization directions than learning from raw data. Accordingly, we design three types of mathematical metrics to evaluate feature representations of the DNN. In experiments, we diagnosed various DNNs, and above hypotheses were verified.