Almost-Optimal Best Restless Markov Arm Identification with Fixed Confidence

Learn about identifying the best arm in restless multi-armed bandit systems through this research lecture presented by Dr. P N Karthik from the National University of Singapore. Explore groundbreaking findings on minimizing expected stopping time while maintaining fixed confidence in systems where each arm operates as a homogenous, discrete-time Markov chain. Discover novel approaches to handling unknown transition probability matrices in restless bandits, extending beyond previous work that focused on independent observations and known transition probabilities. Examine the mathematical framework for establishing asymptotic lower bounds on stopping time growth rates and understand how results from Markov Decision Processes contribute to achieving asymptotic optimality. Gain insights into this collaborative research effort with Vincent Tan (NUS), Ali Tajer (RPI), and Arpan Mukherjee (RPI), which advances the field of multi-armed bandit problems and their applications in distributed learning systems.