Reproduce “SelectNAdapt: Support Set Selection for Few-Shot Domain Adaptation” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Youssef Dawoud, Gustavo Carneiro, Vasileios Belagiannis
SelectNAdapt introduces an unsupervised framework for few-shot domain adaptation, intelligently selecting representative target domain samples to form an effective support set. This approach improves model performance by leveraging self-supervised feature adaptation and distance-based clustering, consistently outperforming random selection across multiple benchmarks like PACS, Office-31, and VisDA.
Key insights
- Intelligent, representation-driven selection of support samples yields superior few-shot domain adaptation results compared to random or uncertainty-based selection methods.
- Self-supervised learning on the unlabeled target domain is crucial for learning robust, target-specific features that enable effective clustering and pseudo-label generation.
- Samples closest to cluster centers are more beneficial for correcting Batch Normalization statistics during adaptation than samples near decision boundaries.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.