Federated Linear Contextual Bandit

Federated linear contextual bandits address the challenge of collaboratively learning optimal strategies from decentralized data while preserving user privacy. Research focuses on developing algorithms that efficiently handle heterogeneous client data and incorporate differential privacy mechanisms, such as user-level central and local differential privacy, to protect sensitive information. Key algorithmic approaches include adaptations of existing centralized bandit algorithms like LinUCB and LinTS, often incorporating clustering techniques for heterogeneous clients or novel encryption methods for vertical federated settings. This field is significant for enabling privacy-preserving machine learning in distributed environments, with applications ranging from personalized recommendations to online advertising.