Core Concepts
Active learning method SUPClust focuses on identifying points at decision boundaries to improve model performance.
Abstract
Active learning optimizes model performance by selecting informative data points.
SUPClust targets points at decision boundaries for model refinement.
Self-supervised representation learning and clustering are used to identify relevant points.
Experimental results show strong model performance and improvement in scenarios with class imbalance.
SUPClust addresses the "cold start problem" and performs well in low-budget regimes.
Ablation study confirms the necessity of all components in SUPClust.
Results show robust performance of SUPClust compared to baseline strategies.
SUPClust demonstrates strong performance in imbalanced settings.
Utilizing pre-trained embeddings enhances performance across datasets.
Diversity-based methods perform better in low-budget regimes.
SUPClust provides a non-label-based means of quantifying sample distance to decision boundaries.
SUPClust contributes to understanding active learning dynamics.
Stats
labeling these points leads to strong model performance.
improvement is observed even in scenarios characterized by strong class imbalance.
data distributions often include outliers.
Quotes
"Active learning aims to maximize performance by selecting the most informative and valuable data points to be annotated for model training."
"SUPClust avoids the 'cold start problem' by selecting samples close to the decision border between clusters."