AlphaRank: an Artificial Intelligence Approach for Ranking and Selection Problems.

We introduce AlphaRank, an artificial intelligence approach to address thefixed-budget ranking and selection (R S) problems. We formulate the sequentialsampling decision as a Markov decision process and propose a Monte Carlosimulation-based rollout policy that utilizes classic R S procedures as basepolicies for efficiently learning the value function of stochastic dynamicprogramming. We accelerate online sample-allocation by using deep reinforcementlearning to pre-train a neural network model offline based on a given prior. Wealso propose a parallelizable computing framework for large-scale problems,effectively combining "divide and conquer" and "recursion" for enhancedscalability and efficiency. Numerical experiments demonstrate that theperformance of AlphaRank is significantly improved over the base policies,which could be attributed to AlphaRank's superior capability on the trade-offamong mean, variance, and induced correlation overlooked by many existingpolicies.