Description
In many settings, predictions from machine learning models can have life-changing effects on individuals. This course focuses on the theoretical foundations of machine learning models that account for the social impacts of their predictions. Topics we will cover include: Performative Prediction (GMS theorem, performative stability, performative optimality, predictability). Performative Power (Learning vs. steering, ex-ante optimization, ex-post optimization). Algorithmic Collective Action (Signal (un)planting, signal erasing, dynamics of collective action). The Value of Prediction (Identifying the worst-off, allocation, prediction-access ratios). Aggregated Individual Reporting (Reporting-based frameworks, sequential hypothesis testing, post-deployment evaluation). Policy Design (Welfare dynamics, Rawlsian vs. utilitarian policies). Outcome Indistinguishability (Omniprediction, outcome performativity, universal adaptability, multicalibration).