10-880: Game-theoretic probability, statistics and learning

(betting, e-values and martingales --- history, philosophy, and mathematics)


This course ad was created by GPT4 when given my course title and description. It was unable to correct the typos in the title despite my explicit prompting. Given the likely scrambled nature of my class (an experimental first-time offering), its title may in fact be more suitable.

Location/time: DH 1211, Tue/Thu 3:30-4:50pm.

Aaditya Ramdas (CMU)



TAs: Ian Waudby-Smith (iwaudbys@andrew.cmu.edu) and Ben Chugg (benchugg@cmu.edu)

Course description We will study the game-theoretic foundation for probability, statistics and learning, and its applications to prediction, testing and estimation. It will distinctively employ modern concepts around betting, e-values, and martingales, effectively bridging Bayesian, frequentist and model-free or adversarial perspectives on these topics.

Target audience: PhD students who are interested in some subset of these topics.

Prerequisites: Students should ideally already have taken some graduate-level probability, statistics and machine learning. Exceptions can be made for students with strong mathematical backgrounds.

Tentative syllabus



Additional Links

SAVI (safe, anytime-valid inference) workshop --- Jun 28 to Jul 2, 2021 in Eindhoven.