Bias Mitigation
Bias mitigation in machine learning aims to create fairer and more equitable algorithms by addressing biases stemming from training data and model architectures. Current research focuses on developing and evaluating various bias mitigation techniques, including data augmentation strategies (like mixup and proximity sampling), adversarial training methods, and post-processing approaches such as channel pruning and dropout. These efforts span diverse applications, from computer vision and natural language processing to medical image analysis and recommender systems, highlighting the broad significance of this field for ensuring responsible and ethical AI development. The ultimate goal is to improve model fairness without sacrificing accuracy or utility, leading to more equitable outcomes across different demographic groups.
Papers
Identifying and Adapting Transformer-Components Responsible for Gender Bias in an English Language Model
Abhijith Chintam, Rahel Beloch, Willem Zuidema, Michael Hanna, Oskar van der Wal
Co$^2$PT: Mitigating Bias in Pre-trained Language Models through Counterfactual Contrastive Prompt Tuning
Xiangjue Dong, Ziwei Zhu, Zhuoer Wang, Maria Teleki, James Caverlee
Detecting and Mitigating Algorithmic Bias in Binary Classification using Causal Modeling
Wendy Hui, Wai Kwong Lau