登录
洞察 - Bias Mitigation in Language Models
Local Contrastive Editing for Controlling Gender Stereotypes in Language Models
By localizing and editing specific weights in language models, it is possible to control and mitigate encoded gender stereotypes while preserving the model's overall performance.
AXOLOTL: Fairness through Assisted Self-Debiasing of Large Language Model Outputs
AXOLOTL introduces a novel post-processing framework for debiasing Large Language Model outputs, ensuring fairness and performance preservation.
Mitigating Social Bias in Language Models without Demographic Information
DAFAIR proposes a novel approach to mitigate social bias in language models without relying on demographic information, achieving competitive performance while reducing bias.
