A framework where a learner repeatedly chooses actions from a convex set and incurs losses, adapting based on feedback.