Performative Prediction Workshop @ FOCS 2026

The use of algorithmic predictions in decision-making leads to a feedback loop in which the models we deploy actively influence the data distributions we see and later retrain on. For instance, traffic predictions alter traffic patterns, financial forecasts move market prices, and the outputs of today's language models shape the data that future models are trained on.

Performative prediction is a recent framework that ties together ideas from learning theory, game theory, and causality to formalize learning in these dynamic environments, giving rise to new equilibrium notions with their own computational and statistical challenges. While a first wave of work has started to map out basic solution concepts and the convergence behavior of classical ML algorithms in this setting, the area has far more questions than answers.

The goal of this workshop is to introduce the FOCS community to the area, explore connections to neighboring fields such as equilibrium computation and algorithmic game theory, and spark a discussion of new research directions.

Speakers and Workshop Schedule

Talk titles and times are tentative.