Regret of Age-of-Information Bandits

We consider a system with single source that measures/tracks time-varying quantity and periodically attempts to report these measurements monitoring station. Each update from the has be scheduled on one of $K$ available communication channels. The probability success each attempted is function channel used. This unknown scheduler. metric interest Age-of-Information (AoI), formally defined as time elapsed since destination received recent most source. model our scheduling problem variant multi-arm bandit channels arms. characterize lower bound AoI regret achievable by any policy performance UCB, Thompson Sampling, their variants. Our analytical results show UCB sampling are order-optimal for bandits. In addition, we propose novel policies which, unlike use current make decisions. Via simulations, proposed AoI-aware outperform existing AoI-agnostic policies.