Is Label Smoothing Truly Incompatible with Knowledge Distillation: An Empirical Study
This work addresses a practical problem for machine learning practitioners using knowledge distillation by clarifying when label smoothing can be effectively combined with it.
This paper empirically investigates the claim that label smoothing is incompatible with knowledge distillation, finding through extensive experiments on image classification, binary networks, and neural machine translation that the incompatibility is one-sided and imperfect, with label smoothing only losing effectiveness in specific circumstances.
This work aims to empirically clarify a recently discovered perspective that label smoothing is incompatible with knowledge distillation. We begin by introducing the motivation behind on how this incompatibility is raised, i.e., label smoothing erases relative information between teacher logits. We provide a novel connection on how label smoothing affects distributions of semantically similar and dissimilar classes. Then we propose a metric to quantitatively measure the degree of erased information in sample's representation. After that, we study its one-sidedness and imperfection of the incompatibility view through massive analyses, visualizations and comprehensive experiments on Image Classification, Binary Networks, and Neural Machine Translation. Finally, we broadly discuss several circumstances wherein label smoothing will indeed lose its effectiveness. Project page: http://zhiqiangshen.com/projects/LS_and_KD/index.html.