AI Paper Atlas

Research map / Label-efficient and robust learning / Zero-/few-shot and meta-learning

EASE: Unsupervised Discriminant Subspace Learning for Transductive Few-Shot Learning

Hao Zhu, Piotr Koniusz

CVPR 2022 · Label-efficient and robust learning › Zero-/few-shot and meta-learning · approach: Alignment-based few-shot metric learning · 58 citations (OpenAlex)

Check 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
Learn the EASE discriminant subspace and use SIAMESE clusteringResNet-12 few-shot inference · mini-ImageNet, transductive 1-shotaccuracy: 67.19 ± 0.55 -> 70.47 ± 0.30Improved
Learn the EASE discriminant subspace and use SIAMESE clusteringResNet-12 few-shot inference · tiered-ImageNet, transductive 1-shotaccuracy: 73.21 ± 0.58 -> 84.54 ± 0.27Improved
Use EASE+SIAMESE instead of the strongest listed methodWRN-28-10 few-shot inference · mini-ImageNet, transductive 1-shotaccuracy: 83.05 ± 0.79 -> 83.00 ± 0.21No clear effect
Use EASE+SIAMESE instead of iLCTWRN-28-10 few-shot inference · tiered-ImageNet, transductive 1-shotaccuracy: 88.50 ± 0.75 -> 88.96 ± 0.23Improved
Add EASE to QR+SIAMESE (ablation)ResNet-12 few-shot inference · CIFAR-FS, transductive 1-shotaccuracy: 75.46±0.26 -> 78.41±0.29Improved
Replace SIAMESE with Soft K-means (ablation)EASE few-shot inference · mini-ImageNet, transductive 1-shot, ResNet-12accuracy: 70.47 ± 0.30 -> 57.00 ± 0.26Hurt
Replace EASE with QR matrix decomposition (ablation)SIAMESE few-shot inference · mini-ImageNet, transductive 1-shot, ResNet-12accuracy: 70.47 ± 0.30 -> 68.66 ± 0.37Hurt
Use EASE+SIAMESE rather than Soft K-means for class estimation (ablation)DenseNet few-shot inference · mini-ImageNet, transductive 1-shotaccuracy: 74.30 ± 0.26 -> 79.42 ± 0.27Improved
Add SIAMESE's support-label constraints to Sinkhorn K-means (ablation)EASE features · few-shot benchmarksaccuracy: 0.3–0.4%Improved
Use the proposed EASE+SIAMESE inferenceResNet-12 few-shot inference · mini-ImageNet, semi-supervised 1-shotaccuracy: 70.99±0.91 -> 73.90±0.29Improved
Use the proposed EASE+SIAMESE inferenceWRN-28-10 few-shot inference · tiered-ImageNet, semi-supervised 1-shotaccuracy: 89.35±0.68 -> 90.08±0.62Improved
Increase the number of query samples (ablation)EASE transductive inference · mini-ImageNetaccuracyImproved

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