Advances in Active Learning

The field of active learning is moving towards developing more efficient and effective methods for selecting the most informative samples for labeling, with a focus on reducing annotation costs and improving model performance. Researchers are exploring new strategies for active learning, including the use of predictive models, sample-aware dynamic soft prompts, and spatial diversity. These innovations have the potential to significantly improve the accuracy and efficiency of active learning systems. Notable papers in this area include: PromptAL, which proposes a hybrid active learning framework that accounts for the contribution of each unlabeled data point in aligning the empirical distribution with the target distribution. Exploring Active Learning for Semiconductor Defect Segmentation, which demonstrates the effectiveness of active learning in reducing annotation costs for semantic segmentation tasks in the context of semiconductor defect segmentation.

Sources

To Label or Not to Label: PALM -- A Predictive Model for Evaluating Sample Efficiency in Active Learning Models

Experimenting active and sequential learning in a medieval music manuscript

PromptAL: Sample-Aware Dynamic Soft Prompts for Few-Shot Active Learning

Exploring Active Learning for Label-Efficient Training of Semantic Neural Radiance Field

Exploring Active Learning for Semiconductor Defect Segmentation

Exploring Spatial Diversity for Region-based Active Learning

Learning from Hard Labels with Additional Supervision on Non-Hard-Labeled Classes

Built with on top of