The TDAI CoP on Foundations of Data Science and AI will host guest speaker Professor Weijie Su, for his presentation on How Statistics Can Advance Large Language Models: Fairness Alignment and Watermarking. Su is an Associate Professor of Statistics and Data Science, Associate Professor of Computer and Information Science, as well as Co-director of the Penn Research in Machine Learning at the University of Pennsylvania.
About the talk
Large language models (LLMs) have rapidly emerged as a transformative innovation in machine learning. However, their increasing influence on human decision-making processes raises critical societal questions. In this talk, we will demonstrate how statistics can help address two key challenges: ensuring fairness for minority groups through alignment and combating misinformation through watermarking. First, we tackle the challenge of creating fair LLMs that equitably represent and serve diverse populations. We derive a regularization term that is both necessary and sufficient for aligning LLMs with human preferences, ensuring equitable outcomes across different demographics. Second, we introduce a general statistical framework to analyze the efficiency of watermarking schemes for LLMs. We develop optimal detection rules for an important watermarking scheme recently developed at OpenAI and empirically demonstrate its superiority over the existing detection method.
Throughout the talk, we will showcase how statistical insights can not only address pressing challenges posed by LLMs but also unlock substantial opportunities for the field of statistics to drive responsible generative AI development. This talk is based on arXiv:2405.16455 and arXiv:2404.01245.
About the Speaker
Professor Weijie Su is an Associate Professor at the Wharton Statistics and Data Science Department, and by courtesy, Departments of Computer and Information Science, Mathematics, and Biostatistics, Epidemiology and Informatics. Su is also a Co-director at the Penn Research in Machine Learning at the University of Pennsylvania, as well as an editorial board member of numerous journals such as the Journal of Machine Learning Research, and the Journal of the American Statistical Association. Weijie Su's research interests include Long-term Mathematical, compute-light approached to understanding deep learning and AI, as well as current interests in Statistical Foundations of Large Language Models, privacy preserving machine learning, high-dimensional statistics, and mathematical optimization. To learn more about the speaker, please follow to his personal website.
