量子强化学习技术及研究进展

Review of quantum reinforcement learning

近年来,强化学习理论和算法研究迅速发展,并且在竞争博弈、智能控制、分析预测、优化调度等领域得到广泛应用.但是,传统强化学习算法学习效率低、系统开销大,尤其是面对复杂任务时这种情况更为严重.结合量子计算特性,可实现对强化学习算法的加速,由此提出的量子强化学习技术,对强化学习技术的发展赋予了全新的动力与广阔的前景,引发了日益广泛的关注.文章对量子强化学习技术及其研究进展进行了介绍、分析与展望.首先,分别对量子计算和强化学习的基本概念和原理进行了介绍.在此基础上,介绍了量子强化学习的基本思想与机制,并从两方面分析介绍了量子强化学习的研究与进展:①传统计算环境下,将量子特性融入到强化学习以提高算法效率;②量子计算环境下,将经典环境量子化之后,智能体同环境进行量子化交互的强化学习技术.最后,对量子强化学习的应用前景进行了展望.

The recent years have witnessed an explosive development of reinforcement learning(RL),which has great potential in competing games,intelligent controlling,analyzing,predicting,optimizing and scheduling,etc.However,traditional RL generally has the disadvantage of slow convergence with a high system cost,especially facing complicated tasks.Researchers have found that the integration of quantum computing and RL can accelerate the RL algorithms,and proposed quantum reinforcement learning(QRL).This will further promote the development and application of RL effectively.In this paper,we make a comprehensive introduction and analysis on state-of-the-art QRL technology.Firstly,we introduce the basic concepts and principles of quantum computing and RL respectively.Then,we introduce the basic ideas and schemes of QRL,and analyze its development in the aspects of①integrating quantum characteristics into RL in traditional computing environments,and②RL in a quantum computing environment.Finally,we forecast the potential applications of QRL in the future.