Multiagent Low-Dimensional Linear Bandits

We study a multiagent stochastic linear bandit with side information, parameterized by an unknown vector $\theta ^* \in \mathbb {R}^{d}$ . The information consists of finite collection low-dimensional subspaces, one which contains ^*$ In our setting, agents can collaborate to reduce regret sending recommendations across communication graph connecting them. present novel decentralized algorithm, where communicate subspace indices each other and agent plays projected variant LinUCB on the corresponding (low dimensional) subspace. By distributing search for optimal users learning in subspace, we show that per-agent finite-time is much smaller than case when do not communicate. finally complement these results through simulations.