Risk-Sensitive Bandits - Arm Mixtures Optimality and Regret-Efficient Algorithms

Explore a technical lecture on risk-sensitive multi-armed bandit problems presented by Dr. Arpan Mukherjee from Imperial College London. Delve into a new framework for risk-aware sequential decision-making that unifies various risk measures through distortion riskmetrics. Learn about the groundbreaking observation that optimal strategies often require selecting arm mixtures rather than single arms, challenging conventional bandit algorithms. Discover newly developed algorithms designed to track optimal mixtures when risk measures favor them, and understand the technical challenges in establishing information-theoretic lower bounds for regret under mixtures-optimality settings. Examine open questions and future research directions in risk-sensitive decision-making, guided by Dr. Mukherjee, whose expertise spans signal processing, statistics, and machine learning, developed through his academic journey at Rensselaer Polytechnic Institute and IIT Kharagpur.