Reinforcement learning for inventory management

In order to minimize inventory management costs, a promising route is to utilise a reinforcement learning approach. Indeed, stock management can be modeled as a sequential decision-making process under uncertainty which is often written as a Markov decision process (MDP). In this case, reinforcement learning provides robust solutions for this kind of tasks. Recent progress in machine learning, and RL in particular, involving deep models in other complex domains (deep Q-learning) suggests that the achieving of high-quality results (which can even be transferred from one inventory problem to another via transfer learning, minimizing overheads) even in these types of highly complex environments may be well possible.

A quantum-enhanced method to generate optimal inventory management strategies would bring two main benefits:

• Improved inventory management directly translates in reduction of expenses;
• Better performance ensures more timely delivery of items to customers, and avoids delays.

Our publications and deliverables

• Parametrized Quantum Policies for Reinforcement Learning (NeurIPS 2021)
• Quantum agents in the Gym: a variational quantum algorithm for deep Q-learning (Quantum 6, 720, 2022)
• D5.2 Specification of QRL algorithm for inventory management
• Unsupervised strategies for identifying optimal parameters in Quantum Approximate Optimization Algorithm (EPJ Quantum Technology)
• Equivariant quantum circuits for learning on weighted graphs (npj Quantum Information)
• Quantum Machine Learning Beyond Kernel Methods (The Journal of Chemical Physics)
• D5.5 Implementation of QRL algorithm on real architecture (report)
• High Dimensional Quantum Machine Learning With Small Quantum Computers (Quantum)
• Reinforcement learning assisted recursive QAOA (EPJ Quantum Technology)