Machine learning for quantum device tuning and simulation

在纳米尺度上控制材料已经足够精细，因此需要新的方法来制造这些结构并调整它们的性能。在许多情况下，我们已经达到了人工控制流程的能力极限，当关键的考虑因素是技术应用程序的可扩展性时，更是如此。该项目确定了量子设备，机器学习将为加速、扩展和开放新技术提供关键。贝叶斯优化是最先进的（在时间或金钱上）昂贵的问题。给定可用的数据，贝叶斯优化方法建立了一个低成本的机器学习模型（实际实验的低成本概率近似值），其预测计算成本低。该模型为下一次测量提供最佳位置。机器使用该模型以最具成本效益的方式找到目标。目标是将贝叶斯优化与其他机器学习方法（如神经网络）相结合，以控制量子硬件在环系统。硬件在环测试允许实际的被测硬件与软件模型实时连接。将单个量子设备一个接一个地交换出来，测试结果可以用来模拟在集成设备组件之间传递的真实（因此是不完美的）信号。实验将在低温下用专用射频电子设备实现。该研究方法的新颖之处在于使用硬件在环模拟，使机器能够学习多量子器件电路中的调谐技术。我们的可扩展模拟和测试平台提供了一个硬件在环设施，用于将机器学习应用于模拟环境中的子系统，以加速先进复杂量子技术的开发和原型设计。这个项目是释放量子技术潜力的关键，它允许对量子设备进行快速测量，从而有可能控制技术上相关的量子电路。与巴塞尔大学的合作为我们提供了实验所需的量子设备，与牛津大学工程系的合作使我们能够从致力于人工智能研究的计算机科学家的专业知识中受益，特别是贝叶斯优化和其他机器学习技术，这些技术不需要大量数据，而量子设备则无法获得这些数据。该项目属于EPSRC量子技术研究领域。

Controlling materials on the nanoscale is now sufficiently refined that new methods are needed for fabricating the structures and tuning their performance. In many cases we are reaching the limits of our ability to do this using human control of the process, all the more so when a key consideration is scalability for technological applications. The project identifies quantum devices where machine learning will provide the key to speeding up, scaling up, and opening up new technologies. Bayesian optimisation is state-of-the-art for expensive (in time or money) problems. Given the available data, Bayesian optimisation methods builds a low-cost machine learning model (an inexpensive probabilistic approximation of the actual experiment) whose predictions are cheap to compute. The model leads to the optimal location for the next measurement. The machine uses the model to find an objective in the most cost-effective way.The objective is to combine Bayesian optimisation with other machine learning methods, such as neural networks, to control quantum hardware-in-the-loop systems. Hardware-in-the-loop testing allows actual hardware under test to be interfaced with a software model in real time. Swapping out individual quantum devices one by one, the test results can be used to simulate realistic (and therefore imperfect) signals passing between components of the integrated device. The experiments will be realized at cryogenic temperatures with dedicated radio-frequency electronics. The novelty of the research methodology resides in the use of hardware-in-the-loop simulations to enable the machine to learn tuning techniques in multi quantum device circuits. Our extensible simulation and test platform provides a hardware-in-the-loop facility for applying machine learning to subsystems in a simulated environment, with a view to accelerating the development and prototyping of advanced complex quantum technologies.This project is key to unleash the potential of quantum technologies by allowing fast measurement of quantum devices, and thus the possibility to control technologically relevant quantum circuits. A collaboration with University of Basel provides us with the quantum devices we require for our experiments, and a collaboration with the Department of Engineering at University of Oxford allows us to benefit from the expertise of computer scientist dedicated to the study of artificial intelligence, in particular Bayesian optimisation and other machine learning techniques which do not require large amounts of data, which is not available for quantum devices. This project falls within the EPSRC Quantum technologies research area.