READ : Reinforcement-based Adversarial Learning for Text Classification with Limited Labeled Data

Rohit Sharma, Shanu Kumar, Avinash Kumar · 2025

Read READ : Reinforcement-based Adversarial Learning for Text Classification with Limited Labeled Data by Rohit Sharma, Shanu Kumar, Avinash Kumar (2025) 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-2501-08035.txt.

Description

Rohit Sharma, Shanu Kumar, and Avinash Kumar tackle text classification when labeled examples are scarce. Modern language models sort text well but hunger for thousands of hand-labeled samples, which cost time and money to make. Unlabeled text, by contrast, is cheap to gather. Their method, READ, squeezes value from that unlabeled pool in two ways at once.

A reinforcement learning agent generates varied synthetic text, guided by rewards rather than fixed rules. An adversarial training step then forces the classifier to stay steady when inputs are nudged, which sharpens its judgment. Together the two tricks teach the model from both real labels and manufactured practice data. Across several benchmark datasets, READ beat the strongest existing methods for low-label classification. The authors offer it as a route to good classifiers where labeling budgets are thin.

Catalog record

CREATORRohit Sharma, Shanu Kumar, Avinash Kumar
YEAR2025
COLLECTIONPAPERS
FORMATTXT
FILE SIZE17 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2501-08035

Browse the full library · About the Tower of Babel