Paper Co-authored by MSE Lukasz Golab the Winner of the Best Demo Paper Award at the 2025 International Conference on Extending Data Base Technology
A paper co-authored by Management Science and Engineering professor Lukasz Golab and his Data Science Master’s student Anastasiia Avksientieva at the . This paper proposed a new data-driven method to assess bias in machine learning models. A model is explicitly biased if it is more accurate for some subgroups than others. For example, a biased healthcare model might generate more accurate diagnoses for younger or older individuals. However, even an explicitly unbiased model may be implicitly biased if it is harder for some subgroups to flip the model's decision to a favourable one. For example, what if married individuals whose loan applications were rejected would only need to increase their incomes by an average of ten percent to be approved, but single individuals would need 20 percent higher salaries? In their paper, Golab and co-authors present a software tool that identifies implicit bias in prediction models, toward responsible deployment of AI models in practice.
