AI Paper Atlas

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

POODLE: Improving Few-shot Learning via Penalizing Out-of-Distribution Samples

Duong Hoang Le, Khoi Duc Nguyen, Khoi Nguyen, Quoc-Huy Tran, Rang Nguyen, Binh-Son Hua

NeurIPS 2021 · Label-efficient and robust learning › Zero-/few-shot and meta-learning · approach: Embedding adaptation for few-shot classification · 19 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 POODLE-R to Rot + KDResNet-12 · mini-ImageNet 1-shot, inductiveaccuracy: 65.91 -> 67.80Improved
Add POODLE-B to Rot + KDResNet-12 · mini-ImageNet 1-shot, transductiveaccuracy: 65.91 -> 77.56Improved
Add POODLE-R to Rot + KDResNet-12 · tiered-ImageNet 1-shot, inductiveaccuracy: 69.43 -> 70.47Mixed
Add POODLE-B to Rot + KDResNet-12 · CUB 1-shot, transductiveaccuracy: 79.70 -> 89.88Improved
Use target-domain noisy negatives with POODLE-BRot + KD ResNet-12 · iNat 1-shot, cross-domainaccuracy: 48.90 -> 50.62Improved
Use target-domain noisy negatives with POODLE-BRot + KD ResNet-12 · ISIC 1-shot, cross-domainaccuracy: 32.35 -> 33.56Improved
Use uniform random features instead of base-class negatives (ablation)POODLE, mini-ImageNet 1-shot · in-domain inductive FSLaccuracy: $66.32\pm0.20$ -> $68.20\pm0.20$Improved
Add the push term using base-class negatives (ablation)CE + L_pull, Simple ResNet-12 · mini-ImageNet 1-shot, transductiveaccuracy: 70.91 -> 74.21Improved
Replace base-class negatives with uniform random negatives (ablation)CE + L_pull, Simple ResNet-12 · mini-ImageNet 1-shot, transductiveaccuracy: 70.91 -> 74.82Improved
Remove the pull term and use only the push term (ablation)Rot + KD ResNet-12 · mini-ImageNet 1-shot, inductiveaccuracy: $66.32\pm0.20$ -> $66.32\pm0.20$No clear effect
Fine-tune a (k+1)-way classifier with OOD as an extra class (ablation)Rot + KD ResNet-12 · mini-ImageNet 1-shot, inductiveaccuracy: $66.32\pm0.20$ -> $66.36\pm0.20$No clear effect
Apply label smoothing (ablation)Rot + KD ResNet-12 · mini-ImageNet 5-shot, inductiveaccuracy: $82.99\pm0.13$ -> $80.81\pm0.13$Hurt

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