Metalearning Linear Bandits by Prior Update

Amit Peleg, Naama Pearl, Ron Meir

Fully Bayesian approaches to sequential decision-making assume that problem parameters are generated from a known prior. In practice, such information is often lacking. This proble…