Q-CALC (Quantum Contextual Artificial intelligence for Long-range Correlations)

Q-CALC（用于远程关联的量子上下文人工智能）

• 批准号: 10085547
• 金额: $ 14.67万
• 依托单位: COLDQUANTA UK LIMITED
• 依托单位国家: 英国
• 项目类别: Small Business Research Initiative
• 财政年份: 2023
• 资助国家: 英国
• 起止时间: 2023 至 无数据
• 项目状态: 已结题
• 来源: https://gtr.ukri.org/projects?ref=10085547
• 关键词: CALC; Quantum; Contextual; Artificial; intelligence

Analysis of complex data sets is of practical interest to the UK government. Classical machine learning (ML), and in particular generative modeling, offers the ability to extract useful representations of the phenomena underlying these empirical data. Current state-of-the-art neural network models, including transformer-based architectures such as (Chat)GPT, struggle to accurately analyse data with long-range dependencies, i.e., with a large context for any given data point. This especially characterises data sets of national security interest, including: super-resolution image processing (e.g., threat detection in low-light, high-noise multi-frame satellite images), natural language processing (e.g., efficient intelligence processing), AND automated real-time analysis of time series sensor data (e.g., anomaly detection, failure prediction). Quantum machine learning (QML) models provide an opportunity to address the challenge of long-range correlations in complex data sets, via equipping ML with the power of quantum contextuality: the underlying nature of a quantum system whereby a measurement outcome depends on the full context of preceding measurement outcomes. Contextuality is a core principle by which the reality of quantum systems eclipse classical models. Indeed, the 2022 Nobel Prize in Physics was awarded to experimental verification of quantum phenomena emerging from the principle of contextuality. We propose the Quantum Contextual Artificial intelligence for Long-range Correlations project (Q-CALC) to develop a QML model that leverages quantum contextuality. Our project will leverage modest quantum resource requirements achieve a large speedup over state-of-the-art classical ML models that are fundamentally limited by comparably shallow context.The Q-CALC project primarily addresses three interlocking technical challenges: (1) enhancing the data-processing capability of state-of-the-art classical ML algorithms by incorporating quantum contextuality, (2) achieving accurate characterisation of complex data sets with long-range dependencies, (3) integrating this technology into real-world data analysis workflows to provide impactful solutions within the defence sector. CQUK is particularly well-suited to tackle these challenges, given significant IP in the quantum space of algorithms, software, and broader technology, as well as our existing relationships with multiple UK governmental security and defence organisations.

对复杂数据集的分析对英国政府具有实际意义。经典机器学习（ML），特别是生成式建模，提供了提取这些经验数据背后现象的有用表示的能力。目前最先进的神经网络模型，包括基于变压器的架构，如（Chat）GPT，难以准确分析具有长期依赖关系的数据，即任何给定数据点的大上下文。这尤其体现在与国家安全利益相关的数据集上，包括：超分辨率图像处理（例如，低光、高噪声多帧卫星图像中的威胁检测）、自然语言处理（例如，高效情报处理）和时间序列传感器数据的自动实时分析（例如，异常检测、故障预测）。量子机器学习（QML）模型提供了一个解决复杂数据集中远程相关性挑战的机会，通过为ML配备量子上下文的力量：量子系统的基本性质，其中测量结果依赖于先前测量结果的完整上下文。情境性是量子系统的现实掩盖经典模型的核心原则。事实上，2022年诺贝尔物理学奖被授予了对情境性原理产生的量子现象的实验验证。我们提出了量子上下文人工智能远程关联项目（Q-CALC）来开发一个利用量子上下文的QML模型。我们的项目将利用适度的量子资源需求，实现比最先进的经典ML模型的大幅加速，这些模型基本上受到相对较浅的上下文的限制。Q-CALC项目主要解决三个相互关联的技术挑战：(1)通过结合量子上下文来增强最先进的经典ML算法的数据处理能力，(2)实现具有远程依赖关系的复杂数据集的准确特征，(3)将该技术集成到现实世界的数据分析工作流程中，为国防部门提供有影响力的解决方案。CQUK特别适合应对这些挑战，考虑到算法、软件和更广泛技术的量子领域的重要知识产权，以及我们与多个英国政府安全和国防组织的现有关系。