Abstract
In various applications such as smart grids, the online player is allowed a limited number of switches among decisions. Additionally, real-world scenarios often involve feedback delays or access to near-future predictions. Motivated by this, we study Online Convex Optimization with a switching limit, incorporating feedback delays and predictions. In this extended abstract, we established a near-optimal regret of O(T/S) for delayed feedbacks and a bound of O(T/S - t ) for predictions of t rounds even though the player is only allowed to move at most S times, in expectation, across T rounds. We developed an algorithm which achieves the bounds in both cases and still works when there are both delays and predictions.
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have