Weijie Su
Professor, University of Pennsylvania
Title: Addressing Bias and Copyright in Generative AI: Preference Matching and Fair Compensation
Abstract
Generative AI has demonstrated impressive capabilities in a range of data science and machine learning tasks, but their societal impact raises critical questions. This talk delves into two pressing issues: ensuring fairness in AI systems to adequately represent minority groups and addressing the challenges of using copyrighted data for training these models. We introduce Preference Matching (PM) RLHF, a new approach to mitigating algorithmic bias in reinforcement learning from human feedback, demonstrating improvements in aligning AI outputs with diverse human preferences. PM RLHF leverages a novel regularizer derived from an ordinary differential equation to balance response diversity and reward maximization. Additionally, we propose an economic framework that fairly compensates copyright owners based on their proportional input, leveraging probabilistic modeling and cooperative game theory. This framework fosters a sustainable and equitable data-sharing ecosystem, promoting mutual benefits for both AI developers and copyright owners through enhanced collaboration and fair compensation.
About Weijie Su
Weijie Su is an Associate Professor at the University of Pennsylvania, where he holds appointments in the Wharton Statistics and Data Science Department and the Department of Computer and Information Science. He is a co-director of the Penn Research in Machine Learning Center. Prior to joining Penn, he received his Ph.D. from Stanford University in 2016 and his bachelor’s degree from Peking University in 2011. His research interests span privacy-preserving machine learning, theoretical aspects of large language models, foundations of learning theory, high-dimensional statistics, and mathematical optimization. He serves as an associate editor of Foundations and Trends in Machine Learning, Operations Research, Journal of the American Statistical Association, and the Journal of Machine Learning. He is a recipient of the Stanford Theodore Anderson Dissertation Award, an NSF CAREER Award, a Sloan Research Fellowship, the IMS Peter Gavin Hall Prize, the SIAM Early Career Prize in Data Science, the ASA Gottfried Noether Early Career Award, and the ICBS Frontiers of Science Award in Mathematics.