Timezone: »

 
Poster
What Makes a "Good" Data Augmentation in Knowledge Distillation - A Statistical Perspective
Huan Wang · Suhas Lohit · Michael Jones · Yun Fu

Tue Nov 29 09:00 AM -- 11:00 AM (PST) @ Hall J #706
in Poster Session 1 »

Knowledge distillation (KD) is a general neural network training approach that uses a teacher model to guide the student model. Existing works mainly study KD from the network output side (e.g., trying to design a better KD loss function), while few have attempted to understand it from the input side. Especially, its interplay with data augmentation (DA) has not been well understood. In this paper, we ask: Why do some DA schemes (e.g., CutMix) inherently perform much better than others in KD? What makes a "good" DA in KD? Our investigation from a statistical perspective suggests that a good DA scheme should reduce the covariance of the teacher-student cross-entropy. A practical metric, the stddev of teacher’s mean probability (T. stddev), is further presented and well justified empirically. Besides the theoretical understanding, we also introduce a new entropy-based data-mixing DA scheme, CutMixPick, to further enhance CutMix. Extensive empirical studies support our claims and demonstrate how we can harvest considerable performance gains simply by using a better DA scheme in knowledge distillation. Code: https://github.com/MingSun-Tse/Good-DA-in-KD.

Keywords: knowledge distillation · Data Augmentation · CutMix · teacher's mean probability · covariance of teacher-student cross-entropy · CutMixPick

Author Information

Huan Wang
Suhas Lohit (Mitsubishi Electric Research Labs)
Michael Jones (MERL)
Yun Fu (Northeastern University)

More from the Same Authors