Alternatives all have fixed probabilities for success. These problems can be modelled after deciding between Bernoulli urns with different proportions of white and gray balls.

University of Michigan - Ann Arbor

Problems which add uncertainty to learning the best alternative. Rewards for potential alternatives are represented by distributions (not fixed outcomes). Bandit problems can model pharmaceutical drug trials, choice among technologies, where to place advertisements and anything else with uncertain payoffs. Can help figure out optimal explore-exploit tradeoff.

Learn Before

Related