Our new X account is live! Follow @wizwand_team for updates
WorkDL logo mark

Knowledge Distillation with the Reused Teacher Classifier

About

Knowledge distillation aims to compress a powerful yet cumbersome teacher model into a lightweight student model without much sacrifice of performance. For this purpose, various approaches have been proposed over the past few years, generally with elaborately designed knowledge representations, which in turn increase the difficulty of model development and interpretation. In contrast, we empirically show that a simple knowledge distillation technique is enough to significantly narrow down the teacher-student performance gap. We directly reuse the discriminative classifier from the pre-trained teacher model for student inference and train a student encoder through feature alignment with a single $\ell_2$ loss. In this way, the student model is able to achieve exactly the same performance as the teacher model provided that their extracted features are perfectly aligned. An additional projector is developed to help the student encoder match with the teacher classifier, which renders our technique applicable to various teacher and student architectures. Extensive experiments demonstrate that our technique achieves state-of-the-art results at the modest cost of compression ratio due to the added projector.

Defang Chen, Jian-Ping Mei, Hailin Zhang, Can Wang, Yan Feng, Chun Chen• 2022

Related benchmarks

TaskDatasetResultRank
Image ClassificationCIFAR-100 (test)
Accuracy78.59
3518
Domain GeneralizationVLCS
Accuracy83.92
238
Image ClassificationImageNet (test)
Top-1 Acc77.6
235
Domain GeneralizationDomainBed (OH, TI, VLCS, PACS, DN) (test)
Accuracy (OH)76.89
33
Image ClassificationCIFAR-100 1.0 (val)
Top-1 Acc78.08
18
Domain GeneralizationOfficeHome DomainBed (OOD)
Avg OOD Accuracy66.76
16
Retinal Disease ClassificationTOPCON-MM
MAP65.1
11
ClassificationBCIC-2A 9 subjects
Weighted F115.98
10
ClassificationBCIC-2B 9 subjects
AUROC57.91
10
Movement RegressionFALCON-M1
R2 Score37.3
9
Showing 10 of 10 rows

Other info

Follow for update