Quantum Machine Learning for Fraud Detection

金融欺诈和未经授权的支付是对数字经济的威胁。随着交易率每年增长5%，英国每天进行6030万笔支付卡交易，准确的欺诈检测需要分析大型数据集。在2019年8290亿英镑的信用卡交易总额中，有6.206亿英镑因欺诈和未经授权的操作而损失[英国金融，欺诈事实2020报告]。目前的欺诈预防率估计超过62%。为尽量减少网络威胁的发生率和影响，必须改善这一点。汇丰在各业务部门使用人工智能（AI）来改善运营，提供数据驱动的预测，提高客户满意度，并检测金融欺诈。后者要求分析财务数据，以查明和标记异常和潜在的欺诈活动。由于欺诈者所采用的策略随时间而变化，因此需要在没有关于正常（名义）和异常（欺诈）交易的先验知识的情况下进行检测。在这个ISCF Germinator项目中，汇丰将与埃克塞特大学合作开发用于异常检测的量子计算协议，以应对这一挑战：利用量子计算方法推进最先进的无监督机器学习（ML）方法。作为概念验证，我们将对这些方法进行测试，评估具有量子资源的无监督机器学习的能力，并估计其未来实施的时间轴。我们的目标是开发一个量子支持的解决方案，该解决方案将显著减少和防止欺诈，同时建立在当前最先进的机器学习解决方案的基础上。

Financial fraud and unauthorised payments represent a threat to the digital economy. With the annual increase of 5% in transaction rates, and 60.3 million payment card transactions performed in the UK every day, accurate fraud detection requires the analysis of large datasets. For the card transaction total of £829 billion in 2019, £620.6 million were lost due to fraudulent and unauthorized operation \[UK Finance, Fraud the Fact 2020 report\]. The current fraud prevention rate is estimated to beat 62% percent. To minimise the incidence and impact of cyber threats, this must be improved.HSBC uses artificial intelligence (AI) in various branches of business to improve operations, offer data-driven predictions, increase customer satisfaction, and detect financial fraud. The latter requires analysing financial data to identify and flag the unusual and potentially fraudulent activity. As the strategies employed by fraudsters change over time, the detection needs to be performed without prior knowledge about normal (nominal) and abnormal (fraudulent) transactions. In this ISCF Germinator project, HSBC will partner with the University of Exeter to develop quantum computing protocols for anomaly detection to address this challenge: advancing the state-of-the-art unsupervised Machine earning (ML) methods with quantum computing approaches. Testing these methods as a proof-of-concept, we will assess the power of unsupervised ML with quantum resources and estimate the timeline for their future implementation.Our goal is to develop a quantum-enabled solution which will significantly reduce and prevent fraud, whilst building on current state-of-the-art ML solutions.