Partial Monitoring

Partial monitoring addresses sequential decision-making problems where feedback on actions is incomplete or delayed, aiming to minimize cumulative losses despite this limited information. Current research focuses on developing algorithms, such as those based on follow-the-regularized-leader and information-directed sampling, that achieve optimal or near-optimal regret bounds in both stochastic and adversarial environments, often incorporating neural networks for improved performance. These advancements have implications for various applications, including active learning, online advertising, and structural health monitoring, by enabling efficient learning from partially informative data streams.