AI Paper Atlas

Research map / Label-efficient and robust learning / Semi-supervised learning

Automatically Discovering and Learning New Visual Categories with Ranking Statistics

Kai Han, Sylvestre-Alvise Rebuffi, Sebastien Ehrhardt, Andrea Vedaldi, Andrew Zisserman

ICLR 2020 · Label-efficient and robust learning › Semi-supervised learning · approach: Holistic semi-supervised pipelines · 34 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
Add rank-statistics pairwise pseudo-labeling with BCE training (ablation)Novel category discovery · CIFAR-10ACC: 26.2 -> 90.4Improved
Add consistency constraints (ablation)Novel category discovery · CIFAR-10ACC: 82.6 -> 90.4Improved
Add cross-entropy training on labelled data (ablation)Novel category discovery · CIFAR-10ACC: 84.7 -> 90.4Improved
Add self-supervised pretraining (ablation)Novel category discovery · CIFAR-10ACC: 89.4 -> 90.4Improved
Add incremental learning of discovered classes (ablation)Ours full · CIFAR-10ACC: 90.4 -> 91.7Improved
Add incremental learning of discovered classes (ablation)Ours full · CIFAR-100ACC: 73.2 -> 75.2Improved
Add incremental learning of discovered classes (ablation)Ours full · SVHNACC: 95.0 -> 95.2Improved
Use the full rank-based joint method instead of DTC with self-supervised initializationNovel category discovery · CIFAR-10, CIFAR-100, and SVHNACC: 88.7$\pm$0.3\% -> 90.4$\pm$0.5\%; 67.3$\pm$1.2\% -> 73.2$\pm$2.1\%; 75.7$\pm$0.4\% -> 95.0$\pm$0.2\%Improved
Use the full method instead of DTCNovel category discovery · OmniGlotACC: 89.0\% -> 89.1\%Improved
Use the full method instead of DTCNovel category discovery · ImageNetACC: 78.3\% -> 82.5\%Improved
Replace self-supervised pretraining with metric-learning pretrainingNovel category discovery · OmniGlot and ImageNetACCNo clear effect

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