Robust Policies for Multi-Unit Allocation and Lost-Sales Inventory Using Reinforcement and Online Learning

Abstract

Efficient allocation of multiple units of scarce resources and management of inventory with potential lost sales are fundamental challenges in operations and supply chain management. Traditional methods often assume static demand and full observability, but real-world markets are inherently uncertain and dynamic. This paper explores the design and analysis of robust policies for multi-unit allocation and lost-sales inventory systems through reinforcement learning and online learning frameworks. We provide theoretical insights, practical algorithms, and empirical evaluations to demonstrate that learning-based strategies can adapt to stochastic demand and non-stationary environments while maintaining near-optimal performance.