Achieving Algorithmic Fairness through Label Flipping
Primary Investigator:
Romila Pradhan
Shashank Thandri
Abstract
As machine learning (ML) and artificial intelligence (AI) become increasingly prevalent in high-stake decision making, fairness has emerged as a critical societal issue. Individuals belonging to diverse groups receive different algorithmic outcomes largely due to the inherent errors and biases in the underlying training data, thus resulting in violations of group fairness or bias.
We address the problem of resolving group fairness by flipping the labels of instances in the training data. We propose solutions to obtain an ordering in which the labels of training data instances should be flipped to reduce the bias in predictions of a model trained over the modified data. We experimentally evaluate our solutions on several real-world datasets and demonstrate that bias is reduced by flipping a small fraction of training data labels.