On the Impact of Knowledge Distillation for Model Interpretability
This work addresses the need for more reliable and interpretable models in various fields, though it is incremental as it builds on existing KD research.
The study tackled the problem of whether knowledge distillation (KD) enhances model interpretability, finding that KD improves both accuracy and interpretability by transferring class-similarity information from teacher to student models, as measured by concept detectors in network dissection.
Several recent studies have elucidated why knowledge distillation (KD) improves model performance. However, few have researched the other advantages of KD in addition to its improving model performance. In this study, we have attempted to show that KD enhances the interpretability as well as the accuracy of models. We measured the number of concept detectors identified in network dissection for a quantitative comparison of model interpretability. We attributed the improvement in interpretability to the class-similarity information transferred from the teacher to student models. First, we confirmed the transfer of class-similarity information from the teacher to student model via logit distillation. Then, we analyzed how class-similarity information affects model interpretability in terms of its presence or absence and degree of similarity information. We conducted various quantitative and qualitative experiments and examined the results on different datasets, different KD methods, and according to different measures of interpretability. Our research showed that KD models by large models could be used more reliably in various fields.