QTRL: Toward Practical Quantum Reinforcement Learning via Quantum-Train

QTRL introduces a novel quantum-classical hybrid approach for reinforcement learning using the Quantum-Train method. It overcomes key limitations in quantum reinforcement learning, including:

• No quantum inference: final trained model is classical, ensuring fast, low-latency inference using standard hardware.
• Efficient training: leverages a quantum circuit to reduce classical policy network parameters by a polylogarithmic factor.
• Practical deployment: eliminates data encoding overhead and reduces quantum hardware dependency.

By shifting the quantum computation to the training phase only, QTRL offers a scalable and cost-effective strategy for applying quantum machine learning to real-world decision-making systems.

Recommended citation: Liu, Chen-Yu; Lin, Chu-Hsuan Abraham; Yang, Chao-Han Huck; Chen, Kuan-Cheng; & Hsieh, Min-Hsiu. (2024). "QTRL: Toward Practical Quantum Reinforcement Learning via Quantum-Train." Proceedings of the 2024 IEEE International Conference on Quantum Computing and Engineering (QCE).
Download Paper