以深度強化學習實現抗噪量子閘

Abstract

量子計算在加密、金融、科學模擬等領域革新中深具潛力，然而現實世界中，量子硬體的雜訊會嚴重地妨礙實行量子演算法，因此實現抗噪量子閘是使量子計算發揮成效的重要前提。本文以創新的方法將量子控制問題整合進強化學習框架中，並使用一種稱為近似策略最佳化的強化學習演算法配合深度神經網路，建立出容錯量子計算所需的高保真、抗雜訊的量子閘。

Quantum computing holds immense promise to revolutionize several industries such as cryptography, finance, scientific simulations and so on. However, the real-world application of quantum algorithms is severely hindered by the presence of noise in quantum hardware. Achieving noise-robust quantum gates is an important prerequisite to harness the power of quantum computing. This thesis presents an innovative way to address the challenge by mapping the quantum gate control problem into the reinforcement learning (RL) framework. Utilizing a RL algorithm called proximal policy optimization equipped with deep neural networks, we achieve constructing high-fidelity and robust single-qubit and two-qubit quantum gates in the presence of quasi-static noise, paving the way for fault-tolerant quantum computation.