Does Knowledge Distillation Really Work?
This challenges common assumptions in model compression for machine learning practitioners, indicating incremental insights.
The paper investigates knowledge distillation, showing that students often fail to closely match teacher predictions despite having the capacity, due to optimization issues and dataset effects, and paradoxically finds that closer matching does not always improve generalization.
Knowledge distillation is a popular technique for training a small student network to emulate a larger teacher model, such as an ensemble of networks. We show that while knowledge distillation can improve student generalization, it does not typically work as it is commonly understood: there often remains a surprisingly large discrepancy between the predictive distributions of the teacher and the student, even in cases when the student has the capacity to perfectly match the teacher. We identify difficulties in optimization as a key reason for why the student is unable to match the teacher. We also show how the details of the dataset used for distillation play a role in how closely the student matches the teacher -- and that more closely matching the teacher paradoxically does not always lead to better student generalization.