Best-Arm Identification
A problem framework in machine learning in which an algorithm adaptively samples from multiple candidate arms to identify the arm with the highest mean reward. It is widely applied in recommender systems, A/B testing, and related domains.
Timeline (last 90 days)
A study unifies multi-armed bandit regret, sequential testing, best-arm identification, and instance-dependent limits of Markov decision processes (MDPs) into a single chain of proof.
The common-variance Gaussian arm model is a standard setting in theoretical analysis of BAI: each arm's reward follows a Gaussian distribution with the same variance and an unknown mean.
All Facts (2)
A study unifies multi-armed bandit regret, sequential testing, best-arm identification, and instance-dependent limits of Markov decision processes (MDPs) into a single chain of proof.
50%UnverifiedThe common-variance Gaussian arm model is a standard setting in theoretical analysis of BAI: each arm's reward follows a Gaussian distribution with the same variance and an unknown mean.
50%