Training individually fair ML models with Sensitive Subspace Robustness
This addresses fairness issues in ML systems like resume screening, but it is incremental as it builds on existing individual fairness concepts with a new optimization method.
The paper tackles the problem of training machine learning models to be individually fair by ensuring performance invariance under sensitive perturbations, such as changes to gender or ethnicity, using a distributionally robust optimization approach, and demonstrates effectiveness on tasks susceptible to gender and racial biases.
We consider training machine learning models that are fair in the sense that their performance is invariant under certain sensitive perturbations to the inputs. For example, the performance of a resume screening system should be invariant under changes to the gender and/or ethnicity of the applicant. We formalize this notion of algorithmic fairness as a variant of individual fairness and develop a distributionally robust optimization approach to enforce it during training. We also demonstrate the effectiveness of the approach on two ML tasks that are susceptible to gender and racial biases.