以近似策略最佳化演算法模擬量子閘控制

Abstract

實現高精度和穩固的量子閘是在具有雜訊的中等規模量子設備中實現可靠的電路深度和達到具有糾錯的容錯量子計算之先決條件，由於量子閘的糾錯操作不應引入過多雜訊。近來，新興的強化學習技術結合深度學習，在最佳控制方面表現出巨大的前景。量子控制問題可以映射到強化學習框架中，以產生控制策略並發現我們所不知道的新物理機制並將其利用。在本文中，我們使用深度神經網路以及採用一種廣泛應用於控制問題的強化學習演算法，稱為近似策略最佳化，以實現高保真度量子閘。與傳統的最佳控制方法相比，此種機器學習方法利用智能體在迭代學習過程中實現精確、高效的量子閘控制。本論文試圖為利用深度強化學習技術研究大型抗噪聲量子計算鋪平道路。

Implementing quantum gates with high precision and robustness is a prerequisite for realizing reliable circuit depth in the noisy intermediate-scale quantum (NISQ) devices and achieving error-corrected fault-tolerant quantum computation, as the error correction operations by quantum gates should not introduce more errors. Recently, the emerging reinforcement learning (RL) techniques combined with deep learning (DL) have shown great promise in control optimization. Quantum control problems can be mapped into a reinforcement learning framework that allows the generation of a control policy by the discovery and exploitation of new physical mechanism that we are not aware of. In this thesis, we utilize deep neural networks and adopt a reinforcement learning algorithm widely, applied on control problems and named proximal policy optimization (PPO), to implement the high-fidelity quantum gates. In contrast to traditional optimal control methods, this machine learning (ML) approach enables an intelligence agent to realize precise and efficient quantum gate control during the iterative learning process. This thesis attempts to pave the way to investigate the large-scale noise-resistant quantum computation with deep reinforcement learning techniques.