Our research team publishes at top venues on active learning, annotation quality, and human-AI collaboration.
A framework for efficiently collecting preference data at scale while maintaining annotator agreement above 95%. Published at NeurIPS 2026.
Novel active learning strategies that account for annotator noise and disagreement, reducing labeling costs by 40% on standard benchmarks.
A new approach to multi-annotator aggregation that preserves minority viewpoints while achieving robust consensus for subjective tasks.
Predicting annotation quality without gold standards using learned embeddings of annotator behavior patterns.
Leveraging annotations from one modality to bootstrap labeling in another, with applications to video and 3D point clouds.
Designing payment structures that incentivize quality while ensuring fair wages across global annotator populations.
We partner with universities and research labs on annotation science. Reach out to discuss opportunities.