AI Paper Atlas

Research map / Label-efficient and robust learning / Knowledge distillation

Robust Active Distillation

Cenk Baykal, Khoa Trinh, Fotis Iliopoulos, Gaurav Menghani, Erik Vee

ICLR 2023 · Label-efficient and robust learning › Knowledge distillation · approach: Rethinking distillation objectives · 2 citations (OpenAlex)

Read the paperCheck your idea against 137,217 papers

AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.

Measured effects reported in the paper

What was changedWhereMeasuredEffect
Use RAD instead of Margin, Uniform, Cluster Margin, Entropy, or UNIXKD for teacher-point selectionSemi-supervised knowledge distillation · CIFAR10, CIFAR100, SVHNtest accuracyImproved
Use RAD instead of the compared selection methodsResNet18 student with ResNet50 teacher · ImageNettest accuracyImproved
Vary the penalty weight w from the default to other evaluated values (ablation)RAD · CIFAR10, CIFAR100, SVHNtest accuracyNo clear effect
Use student prediction entropy instead of margin to define gains (ablation)RAD · CIFAR10, CIFAR100, SVHNtest accuracyNo clear effect
Apply RAD instead of state-of-the-art active-learning selection strategiesStandard active learning · Evaluated active-learning scenariostest accuracyMixed
Use RAD instead of Margin or Uniform across varied model sizes, label-set sizes, query batch sizes, and training epochsKnowledge distillation · SVHNtest accuracyImproved
Select only the highest-gain, lowest-margin points instead of using RAD's robust sampling distribution (ablation)Active distillation · CIFAR10, CIFAR100, SVHNtest accuracyHurt

Closest papers by meaning

Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy