Student: Suyun Liu
Project: Stochastic Multi-Objective Optimization and Fairness in Machine Learning
View: Research Poster (PDF) | Presentation (YouTube)
Department: Industrial and Systems Engineering
Advisor: Luis Vicente
Abstract
Stochastic multi-objective optimization (SMOO) deals with the simultaneous optimization of conflicting stochastic goals. Its numerical solution can be done using the stochastic multi-gradient (SMG) method. We prove that SMG calculates a non-dominated point in the so-called Pareto front at the same rate as classical stochastic descent does for a single objective. The SMG method is framed into a Pareto-front type algorithm for calculating an approximation of the entire Pareto front. One can apply the Pareto-front SMG algorithm to SMOO problems arising from supervised machine learning. For example, in the application of machine learning to real-life decision-making systems like credit scoring and criminal justice, the prediction outcomes might discriminate against people with sensitive attributes, leading to unfairness. We explicitly identify prediction accuracy and fairness in binary classification as two naturally conflicting criteria. Our numerical results demonstrate that the Pareto-front SMG algorithm is capable of determining Pareto fronts with high efficiency and robustness.
About Suyun Liu
Suyun Liu is currently a fourth-year Ph.D. candidate in the Department of Industrial and Systems Engineering at Lehigh University. She is supervised by Prof. Luis Nunes Vicente and mainly working on Stochastic Multi-Objective Optimization and applications in Fair Machine Learning. Before joining Lehigh ISE, she received her master’s degree in the Department of Industrial Engineering and Decision Analysis (IEDA) at The Hong Kong University of Science and Technology and bachelor’s degree from Industrial Engineering at Beijing Jiaotong University.