Speaker: Ms. Yao Yao
From: University of Iowa
Abstract
Artificial intelligence (AI) and machine learning technologies have been used in high-stakes decision making systems such as lending decision, criminal justice sentencing and resource allocation. A new challenge arising with these AI systems is how to avoid the unfairness introduced by the systems that lead to discriminatory decisions for protected groups defined by some sensitive variables (e.g., age, race, gender). I propose new threshold-agnostic fairness metrics and statistical distance-based fairness metrics, which is stronger than many existing fairness metrics in literature. Among the techniques for improving the fairness of AI systems, the optimization-based method, which trains a model through optimizing its prediction performance subject to fairness constraints, is most promising because of its intuitive idea and the Pareto efficiency it guarantees when trading off prediction performance against fairness. I develop new stochastic-gradient based optimization algorithms that leverage the unique structure of the model to expedite the training process with theoretical guarantee. Also, I numerically demonstrate the effectiveness of my approaches on real-world data under different fairness metrics.
For more info, please follow this link.
Read More