Planning and Learning in Risk-Aware Restless Multi-Arm Bandit

Explore the intricacies of risk-aware restless multi-arm bandits in this 48-minute seminar presented by Nima Akbarzadeh from HEC Montréal at the GERAD Research Center. Delve into the generalization of traditional restless multi-arm bandit problems by incorporating risk-awareness into the objective function. Discover the established indexability conditions for risk-aware objectives and learn about a proposed Thompson sampling approach for addressing learning challenges when true transition probabilities are unknown. Gain insights into how this method achieves bounded regret that scales sublinearly with episodes and quadratically with arms. Examine numerical experiments demonstrating the effectiveness of this approach in reducing risk exposure in restless multi-arm bandits.

Planning and Learning in Risk-Aware Restless Multi-Arm Bandit, Nima Akbarzadeh