Papers › Understanding the Role of the Projector in Knowledge Distillation

Understanding the Role of the Projector in Knowledge Distillation

20 Mar 2023arXiv:2303.11098archive 2025-07-28

Roy Miles, Krystian Mikolajczyk

In this paper we revisit the efficacy of knowledge distillation as a function matching and metric learning problem. In doing so we verify three important design decisions, namely the normalisation, soft maximum function, and projection layers as key ingredients. We theoretically show that the projector implicitly encodes information on past examples, enabling relational gradients for the student. We then show that the normalisation of representations is tightly coupled with the training dynamics of this projector, which can have a large impact on the students performance. Finally, we show that a simple soft maximum function can be used to address any significant capacity gap problems. Experimental results on various benchmark datasets demonstrate that using these insights can lead to superior or comparable performance to state-of-the-art knowledge distillation techniques, despite being much more computationally efficient. In particular, we obtain these results across image classification (CIFAR100 and ImageNet), object detection (COCO2017), and on more difficult distillation objectives, such as training data efficient transformers, whereby we attain a 77.2% top-1 accuracy with DeiT-Ti on ImageNet. Code and models are publicly available.

PaperPDFCode

In Syntology Open this paper in Syntology's Atlas, the map of the papers in Syntology's graph and their citations.

Code

roymiles/simple-recipe-distillation officialmentioned in papermentioned on GitHubpytorch report
roymiles/vkd mentioned on GitHubpytorch report
yoshitomo-matsubara/torchdistill mentioned on GitHubpytorch report

Repository list and official/mentioned flags are the archive's, frozen 2025-07-28. Reachability, where shown, is from one Syntology probe window (2026-09-16 to 2026-09-18); repositories not probed show nothing. GitHub stars are not tracked.

Code Syntology ran Syntology

Not run by Syntology. Nothing on this page verifies that the listed code works.

Tasks

Image ClassificationKnowledge DistillationMetric LearningObject Detectionimage-classificationobject-detection

Results from the paper archive 2025-07-28

TaskDatasetModelMetricValueRank at snapshotLeaderboardReport
Knowledge Distillation CIFAR-100 SRD (T:resnet-32x4, S:shufflenet-v2) Top-1 Accuracy (%) 79.86 #1 of 27 Archive leaderboard report
Knowledge Distillation ImageNet SRD (T:RegNety 160 S:DeiT-S) CRD training setting ✘ #12 of 52 Archive leaderboard report
Knowledge Distillation ImageNet SRD (T:RegNety 160 S:DeiT-S) Top-1 accuracy % 82.1 #12 of 52 Archive leaderboard report
Knowledge Distillation ImageNet SRD (T:RegNety 160 S:DeiT-S) model size 22M #12 of 52 Archive leaderboard report
Knowledge Distillation ImageNet SRD (T:RegNety 160 S:DeIT-Ti) CRD training setting ✘ #21 of 52 Archive leaderboard report
Knowledge Distillation ImageNet SRD (T:RegNety 160 S:DeIT-Ti) Top-1 accuracy % 77.2 #21 of 52 Archive leaderboard report
Knowledge Distillation ImageNet SRD (T:RegNety 160 S:DeIT-Ti) model size 6M #21 of 52 Archive leaderboard report
Knowledge Distillation ImageNet SRD (T: ResNet-34 S:ResNet-18) CRD training setting ✓ #38 of 52 Archive leaderboard report
Knowledge Distillation ImageNet SRD (T: ResNet-34 S:ResNet-18) Top-1 accuracy % 71.87 #38 of 52 Archive leaderboard report

Ranks are positions in the archive's leaderboards as they stood at the 2025-07-28 snapshot. Results published since then are not among these rows, so a rank here is not a current standing.

Methods

Knowledge Distillation

Report a problem or propose a change · a person checks every report against the paper or source before anything changes; decisions are listed on /corrections