CRAY Colloquium: Data Valuation in Machine Learning, AI, and Data (MAD) Systems

Valuation in data and AI markets is becoming a cornerstone of data science and artificial intelligence, studied in e-commerce, economics, machine learning, and data management. However, assessing the contribution of individual parties in a coalition completing a task—known as data valuation—remains a significant technical challenge. In this talk, I will present our research on fast, scalable data valuation methods designed for massive AI, machine learning, and data (MAD) systems. Our approach is built on three key innovations: utility- and game-specific evaluation for enhanced accuracy, efficient sampling techniques to reduce computational cost, and Shapley value approximation with limited information to handle extreme scenarios. I will advocate for data valuation as a fundamental service in the next generation MAD systems and explore open challenges and future research directions in this rapidly evolving field.