AI Seminar

Obtaining statistics from selfish agents

Rafael FrongilloPostdoctoral ResearcherMicrosoft Research (NYC)

Scoring rules are payment schemes to elicit a subjective belief, represented as a probability distribution over some disjoint outcomes, given an independent sample from the true distribution. A relatively recent literature explores the more general question of eliciting a statistic, or property, of a distribution. We will go back to the roots of this literature, starting with Savage in 1971, and trace it to the present, culminating in some open problems and preliminary progress for general vector-valued properties. If time allows, we'll also explore some connections to prediction markets and mechanism design.
Rafael Frongillo is a postdoc at MSR-NYC. He earned his Ph.D. at UC Berkeley, advised by Christos Papadimitriou and supported by the NDSEG fellowship. His research lies broadly in algorithmic economics, drawing techniques from game theory, convex analysis, machine learning, and dynamical systems.

Sponsored by