Continual debiasing: A bias mitigation framework for natural language understanding systems

Open Webpage

Mingyu Lee, Junho Kim, Jun-Hyung Park, SangKeun Lee

Published: 2025, Last Modified: 13 Jul 2025Expert Syst. Appl. 2025EveryoneRevisionsBibTeXCC BY-SA 4.0
Abstract: Highlights•Prompt tuning improves the generalization performance of the model.•Learning task features from biased examples boosts the performance.•Preserving important prompt alleviates the trade-off problem of debiasing methods.