ERI: Enhanced Robustness for Approximate Quantum Computing Hardware and Applications

在药理学开发、医疗保健、金融和普通科学等领域的一些计算任务，在经典计算机上需要太长时间才能实用甚至可用。 由于量子力学的奇怪性质，量子计算可能会将这些加速到几分钟甚至几秒钟。然而，量子计算仍处于起步阶段，并面临着诸如量子位（或量子位）数量非常低和噪声水平高等障碍。该项目提出使用机器学习（ML）技术从量子计算机的噪声输出中提取有用的信息。 通过提取干扰输出的噪声的关键特征，ML算法可以将有用信息从噪声中分离出来。这有两个好处，这将是本项目的目标。一方面，它将使量子计算机的使用成为可能，尽管它们具有噪声性质。另一方面，它将允许通过近似简化量子应用，需要更少的硬件资源和计算步骤。通过这种方式，该项目解决了上面提到的两个问题-量子比特数量少和噪声水平高-使量子计算加速成为现实。量子计算作为具有重大社会影响的计算问题的加速器具有巨大的潜力。其中一些问题涉及医疗保健和药物开发、金融或网络安全等领域。尽管近年来在实现实用量子计算方面取得了很大进展，但在其目前有限的量子位数和高错误率下，其真正的潜力无法实现。 这项工作将应用机器学习和统计信号处理概念来增强量子计算应用的鲁棒性。稳健的量子系统将使近似计算方法的应用能够简化量子电路设计、预处理和仿真的硬件和软件需求。 量子电路依赖于多次运行（射击）来找到最终测量状态的解决方案，并以概率分布的形式产生结果。量子计算机的这一独特功能可以转化为一种优势。这项研究是由以下研究假设驱动的：虽然近似量子计算会改变测量的概率分布，但如果这些变化得到很好的理解和应用，它不会损害其正确性。该团队将探索近似电路和软件级近似，以回答以下问题：（i）在多大程度上可以容忍测量结果的概率分布的改变，并且仍然在量子处理中达到正确的结果;（ii）量子近似计算能够在多大程度上减少量子硬件要求和经典预处理要求;（iii）量子近似计算能够在多大程度上减少量子硬件要求和经典预处理要求。（iii）近似技术如何影响结果的概率分布。因此，该项目有两个目标：第一个目标是通过应用机器学习来识别系统结果中的正确解决方案，从而增强量子系统的鲁棒性。利用增强的鲁棒性，第二个目标是识别可以近似的量子计算，更有效地利用硬件资源，同时仍然得到正确的解决方案。该研究的最终目标是为构建强大且硬件高效的量子加速做出贡献。该奖项反映了NSF的法定使命，并通过使用基金会的知识价值和更广泛的影响审查标准进行评估，被认为值得支持。

Some computational tasks in fields such as pharmacological development, healthcare, finances, and general science, take too long on classical computers to be practical or even usable. Quantum Computing can potentially accelerate these to just a few minutes or even seconds thanks to the strange properties of quantum mechanics. However, Quantum Computing is still in its infancy, and facing hurdles such as the very low number of quantum bits (or qubits) and the high levels of noise. This project proposes the use of Machine Learning (ML) techniques to extract useful information out of the noisy outputs of quantum computers. By extracting the key features of the noise that disturbs the output, ML algorithms can separate the useful information from the noise. This has two advantages that will be the goals of this project. On one hand, it will enable the use quantum computers despite their noisy nature. On the other, it will allow for the simplification of the quantum applications through approximations, requiring less hardware resources and computational steps. In this way, the project tackles the two problems mentioned above -low number of qubits and high noise levels- to make quantum computing acceleration a reality. Quantum computing holds great potential as an accelerator of computational problems with significant societal impact. Some of these problems are in the fields of healthcare and drug development, finances or cybersecurity to mention a few. Although great progress has been made in recent years to reach practical quantum computation, its true potential cannot be achieved under its current limited number of qubits and high error rates. This work will apply Machine Learning and Statistical Signal Processing concepts to enhance the robustness of quantum computing applications. Robust quantum systems will enable the application of approximate computing approaches to simplify the hardware and software demands of quantum circuits design, pre-processing and simulation. Quantum circuits rely on multiple runs (shots) to find the solution in the final measured state, and produce their outcome in the form of a probability distribution. This unique feature of quantum computers can be turned into a strength. This research is driven by the following research hypothesis: that although approximate quantum computing will alter the measured probability distribution, it will not compromise its correctness if the changes are well understood and applied. The team will explore approximate circuits and software-level approximation to answer the following questions: (i) to what extent alterations to the probability distribution of the measured outcome can be tolerated, and still reach the right outcome in quantum processing; (ii) to what extent quantum approximate computing is able to reduce the quantum hardware requirements and classical pre-processing demands; (iii) in which way approximation techniques affect the probability distribution of the outcome. Therefore, the project has two objectives: The first objective is to enhance the robustness of quantum systems by applying Machine Learning to identify the correct solutions in the system’s outcome. Taking advantage of the enhanced robustness, the second objective is to identify quantum computations that can be approximated, making more efficient use of hardware resources, while still resulting in correct solutions. The final goal of this research it to contribute to the building of robust and hardware efficient quantum acceleration.This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.