Learning Algorithms for Regenerative Stopping Problems with Applications to Shipping Consolidation in Logistics

We study regenerative stopping problems which involve computing decision making strategies to optimize long-term average cost. Traditional solutions involve estimating the underlying process from data and computing strategies for the estimated model. In this paper, we compare such solutions to learning based solutions, namely, deep reinforcement learning and imitation learning which involve learning a neural network policy from simulations of the underlying process. We evaluate the approaches on a real-world problem of shipping consolidation in logistics and study the tradeoffs between the different approaches