Quantum computing with differentiable quantum transforms

量子计算是一个跨学科的领域，开发一种基于量子力学的新型计算机。存在许多量子算法，能够解决社会和工业相关的问题，跨越一系列领域，这些问题被认为是常规计算机难以解决的。开发算法，特别是那些提供有意义的加速的算法，是非常具有挑战性的。目前，大多数量子编程都是通过用常规语言编写软件来完成的，并通过量子软件库进行增强。量子软件使研究人员能够超越理论，在模拟器和真实的硬件上建立原型和测试算法。然而，好处不仅仅是单向的：正如研究的要求为软件的开发提供信息一样，软件解锁的功能推动了进一步的研究。这项研究是UBC和Xanadu Quantum Technologies之间的合作，Xanadu Quantum Technologies是一家位于多伦多的初创公司，开发了开源软件库PennyLane。PennyLane是专门为量子机器学习设计算法的，通过将自微分应用于量子程序。该软件最近增加了一个名为可微量子变换的新功能。变换修改和操纵量子程序。它们在PennyLane的自动分化框架中的实现已经解锁了训练和学习转换的能力。通过这种合作关系，我们将探索该软件如何帮助我们开发新的方法来编译和优化量子算法，减轻量子硬件中的噪声以及模拟噪声设备。量子计算机在构建和扩展到大尺寸方面具有挑战性;我们今天有许多设备，但它们被认为是小的，并且噪音很大。在这项研究中开发的软件和方法将使我们能够减少运行和测试许多量子算法所需的资源，并在当今的设备上运行时获得更准确的结果。

Quantum computing is an interdisciplinary field developing a new type of computer based on quantum mechanics. A number of quantum algorithms exist that are capable of solving socially- and industrially-relevant problems, across a spectrum of domains, that are considered intractable for regular computers. Developing algorithms, in particular those that provide a meaningful speedup, is extremely challenging. Currently, most quantum programming is done by writing software in regular languages, augmented by quantum software libraries. Quantum software enables researchers to go beyond the theory, and prototype and test algorithms on both simulators and real hardware. However, the benefits are not merely one-way: just as the requirements of research inform the development of software, features unlocked by software drive further research.This research is a partnership between UBC and Xanadu Quantum Technologies, a Toronto-based startup that develops the open-source software library PennyLane. PennyLane is tailored for designing algorithms for quantum machine learning through application of autodifferentiation to quantum programs. The software has recently been augmented with a new feature called differentiable quantum transforms. Transforms modify and manipulate quantum programs. Their implementation within the autodifferentiation framework of PennyLane has unlocked the ability to train and learn transforms. Through this partnership, we will explore how this software can help us develop novel methods for compiling and optimizing quantum algorithms, mitigating noise in quantum hardware, and simulating noisy devices. Quantum computers are challenging to build and to scale to large sizes; we have many devices today, however they are considered small, and are quite noisy. The software and methods developed in this research will enable us to reduce the resources required to run and test many quantum algorithms, as well as obtain more accurate results when running on today's devices.