AI Paper Atlas

Research map / Label-efficient and robust learning / Domain adaptation and generalization

Self-training For Few-shot Transfer Across Extreme Task Differences

Cheng Perng Phoo, Bharath Hariharan

ICLR 2021 · Label-efficient and robust learning › Domain adaptation and generalization · approach: Statistical transfer learning · 17 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
Self-train the source representation on unlabeled target-domain images, retaining teacher pseudo-label groupings and adding SimCLRResNet-10 transfer from miniImageNet · BSCD-FSL 5-way 1-shot, averaged across target datasetsaccuracy: +2.9 points on averageImproved
Add SimCLR self-supervision to STARTUP (ablation)ResNet-10 STARTUP · ISIC 5-way 5-shotaccuracy: 46.48 ± 0.61 -> 47.22 ± 0.61Improved
Add SimCLR self-supervision to STARTUP (ablation)ResNet-10 STARTUP · EuroSAT 5-way 1-shotaccuracy: 62.90 ± 0.83 -> 63.88 ± 0.84Improved
Use unlabeled images from the target domain instead of EuroSAT unlabeled images (ablation)ImageNet-trained ResNet-18 STARTUP · CropDisease 5-way 1-shotaccuracy: 70.40 ± 0.86 -> 85.10 ± 0.74Improved
Remove the base-dataset training loss (ablation)ResNet-10 STARTUP · ISIC 5-way 5-shotaccuracy: 47.22 ± 0.61 -> 45.73 ± 0.61Hurt
Replace SimCLR with rotation self-supervision (ablation)ResNet-10 STARTUP · CropDisease 5-way 1-shotaccuracy: 75.93 ± 0.80 -> 70.53 ± 0.84Hurt
Increase the amount of unlabeled target-domain data (ablation)STARTUP · ISIC few-shot transferaccuracyImproved
Change the student initialization strategy among scratch, teacher initialization, and teacher embedding with a random classifier (ablation)STARTUP without SimCLR · BSCD-FSL datasetsaccuracyMixed
Initialize the student from the teacher and omit SimCLR (ablation)STARTUP · tieredImageNet-less 5-way 1-shotaccuracy: 57.29 ± 0.83 -> 60.39 ± 0.86Improved
Train the student from scratch and omit SimCLR (ablation)STARTUP · tieredImageNet-more 5-way 1-shotaccuracy: 57.68 ± 0.89 -> 63.29 ± 0.90Improved
Apply STARTUP rather than direct transferResNet-10 transfer · miniImageNet 5-way 1-shotaccuracy: 54.18 ± 0.79 -> 54.20 ± 0.81No clear effect
Train STARTUP using unlabeled data from another domain (ablation)ImageNet-trained ResNet-18 STARTUP · CropDisease 5-way 1-shotaccuracy: 70.40 ± 0.86, underperforming naive TransferHurt

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