Label Curation Using Agentic AI

Subhodeep Ghosh, Bayan Divaaniaazar, Md Ishat-E-Rabban, Spencer Clarke, Senjuti Basu Roy · 2026

Read Label Curation Using Agentic AI by Subhodeep Ghosh, Bayan Divaaniaazar, Md Ishat-E-Rabban, Spencer Clarke, Senjuti Basu Roy (2026) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2602-02564.txt.

Description

Subhodeep Ghosh and 4 co-authors coordinate teams of AI agents to produce trustworthy labels without any ground truth, in a 2026 arXiv paper. Labeling is essential for supervised learning, yet human workflows are costly, slow, and uneven as data sets grow in size and kind. Their system, AURA, targets reliability-aware machine annotation at scale.

Multiple agents make and check labels together. At the core sits a classical probabilistic model that jointly infers the hidden true labels and each annotator's reliability through confusion matrices. Expectation-maximization settles conflicting annotations and merges noisy predictions into final labels. Across four benchmark data sets, accuracy improved by up to about six percent over rivals. With poor-quality annotators, gains reached fifty percent. AURA also scores annotator reliability, so label quality can be audited without pre-validation.

Catalog record

CREATORSubhodeep Ghosh, Bayan Divaaniaazar, Md Ishat-E-Rabban, Spencer Clarke, Senjuti Basu Roy
YEAR2026
COLLECTIONPAPERS
FORMATTXT
FILE SIZE33 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2602-02564

Browse the full library · About the Tower of Babel