Decision-making in high-stakes applications, such as healthcare and algorithmic trading, is increasingly data-driven and supported by deep learning models. However, these models are often fragile to slight environmental changes. How can we make them trustworthy? While explainability has been considered a path towards building trustworthy AI models, we argue that it is neither sufficient nor necessary. From a statistical perspective, we propose that algorithms must always generalize to the future consistently by adapting to different conditions. In other words, a trustworthy algorithm should perform well across different conditions. We will discuss some of our recent studies and conclude with key take-away messages.
嘉宾介绍
Qiang Sun is currently a asscociate Professor within the Department of Statistics and Data Sciences and Department of Computer Science at the University of Toronto (UofT) and a visiting faculty member at MBZUAI, leading the StatsLE Lab. He is interested in statistics + AI, with a focus on trustworthy AI, efficient generative AI (GenAI), and next-generation statistics. His research is mostly driven by practical challenges in tech, finance, and science. Prior to his tenure at UofT, he was an associate research scholar at Princeton University. He obtained his PhD from the University of North Carolina at Chapel Hill (UNC-CH) and his BS in SCGY from the University of Science and Technology of China (USTC). In addition to his faculty role, he also serves as an associate editor for Electronic Journal of Statistics (EJS) and as an area chair for several machine learning conferences. He is a James E. Grizzle Distinguished Alumni of UNC-CH, and a receipt of the Connaught Award, and delivers a series of plenary talks.
