A Hierarchical Nearest Neighbour Approach to Contextual Bandits
This work provides an incremental improvement for researchers in online learning and bandit algorithms by enhancing regret performance in specific scenarios.
The paper tackles the adversarial contextual bandit problem in metric spaces by addressing high regret near decision boundaries in prior work, resulting in an algorithm that allows holding out any set of contexts to compute regret while maintaining computational efficiency.
In this paper we consider the adversarial contextual bandit problem in metric spaces. The paper "Nearest neighbour with bandit feedback" tackled this problem but when there are many contexts near the decision boundary of the comparator policy it suffers from a high regret. In this paper we eradicate this problem, designing an algorithm in which we can hold out any set of contexts when computing our regret term. Our algorithm builds on that of "Nearest neighbour with bandit feedback" and hence inherits its extreme computational efficiency.