Adversarial Linear Contextual Bandit

Adversarial linear contextual bandits address the challenge of online decision-making under uncertainty where rewards are determined by an adversary and depend on contextual information. Current research focuses on developing efficient algorithms, such as variations of EXP3 and LinUCB, that minimize regret in this challenging setting, often incorporating techniques to handle delayed feedback, switching costs, and high-dimensional contexts. These advancements are significant for improving the performance of applications like collaborative edge inference and federated learning, where adaptive adversaries and limited feedback are common challenges.