Introduction to Performative Prediction - Tutorial 1

Explore a comprehensive tutorial on performative prediction, a crucial concept in machine learning and social sciences. Delve into the phenomenon where predictions influence their targets, causing distribution shifts in data-generating processes. Learn about self-fulfilling and self-negating predictions, and understand their significance in economics, finance, and digital platforms. Discover the recently established framework for studying performativity in machine learning, including equilibrium notions and optimization challenges. Examine the distinction between learning and steering in performative prediction, and its implications for power dynamics in digital markets. Gain insights into key technical results, drawing connections to statistics, game theory, and causality. Conclude with a discussion on future directions, including the role of performativity in contesting algorithmic systems. Access accompanying slides for a comprehensive learning experience in this 3-hour 10-minute tutorial presented by Celestine Mendler-Duenner and Tijana Zrnic at the Uncertainty in Artificial Intelligence conference.