Collaborative Research: NSF-CSIRO: HCC: Small: Understanding Bias in AI Models for the Prediction of Infectious Disease Spread

合作研究：NSF-CSIRO：HCC：小型：了解预测传染病传播的 AI 模型中的偏差

• 批准号: 2302968
• 负责人: Andreas Zuefle
• 金额: $ 37.37万
• 依托单位: Emory University
• 依托单位国家: 美国
• 项目类别: Standard Grant
• 财政年份: 2023
• 资助国家: 美国
• 起止时间: 2023-04-01 至 2026-03-31
• 项目状态: 未结题
• 来源: https://www.nsf.gov/awardsearch/showAward?AWD_ID=2302968&HistoricalAwards=false
• 关键词: Collaborative; Research; NSF; CSIRO; HCC

Artificial intelligence (AI) provides powerful techniques for understanding and prediction of complex systems such as modeling and predicting the spread of infectious diseases. Despite this, these predictive capabilities are rarely adopted by public health decision-makers to support policy interventions. One of the issues preventing their adoption is that AI methods are known to amplify the bias in the data they are trained on. This is especially problematic in infectious disease models which leverage available large and inherently biased spatiotemporal data. These biases may propagate through the modeling pipeline to decision-making, resulting in inequitable and ineffective policy interventions. This project investigates how the AI disease modeling pipeline can lead from biased data to biased predictions and to derive solutions that mitigate this bias in three aims: 1) creating an AI system to predict the spread of emerging infectious diseases in space and time, 2) simulating a population from which we will collect data often used as input for AI systems in a way that the bias is controlled, and 3) exploring links between bias in the collected data and the resulting bias in the AI model and deriving solutions for their mitigation. The project will enable AI-driven infectious disease models and predictions that will support fair and equitable decision-making and interventions. The project will enrich education and training related to ethical AI practices and will support professional development opportunities for early-career researchers, graduate, undergraduate, and high school students in the United States and Australia. In Aim 1, the team of researchers will use a self-supervised contrastive learning approach that uses mobility prediction as a pre-text task to learn representations of spatial regions. These representations can be used for infectious disease spread prediction given only very little infectious disease ground truth data. The investigators hypothesize that such a model is susceptible to data bias. Thus, in Aim 2, the team of researchers will leverage a large-scale agent-based simulation that will serve as a sandbox world for which we have perfect knowledge of and from which we can collect data and inject various types of bias. For Aim 3, the team of researchers will investigate how different types of simulated data bias leads to biased AI predictions by leveraging different metrics of fairness in AI and studying how these fairness measures can be incorporated into the AI optimization procedure to mitigate bias. By understanding, measuring, and mitigating bias inherent to traditional AI solutions, the project will enable accurate, scalable, and rapid predictions to support fair and equitable decision-making for pandemic prevention.This is a joint project between researchers in the United States and Australia funded by the Collaboration Opportunities in Responsible and Equitable AI under the U.S. NSF and the Australian Commonwealth Scientific and Industrial Research Organization (CSIRO).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.

人工智能（AI）为理解和预测复杂系统（例如建模和预测传染病的传播）提供了强大的技术。尽管如此，公共卫生决策者很少采用这些预测能力来支持政策干预措施。阻止其采用的问题之一是已知AI方法会扩大他们接受过的数据的偏见。这在传染病模型中尤其有问题，这些模型利用可用的大型且固有的偏见时空数据。这些偏见可能通过建模管道传播到决策，从而导致政策干预措施不平等且无效。该项目调查了AI疾病建模管道如何从偏见的数据到偏见的预测，并得出解决这一偏见的解决方案，以三个目标：1）创建一个AI系统，预测一个AI系统在空间和时间上的传播中的传播，2）我们将在该数据中汇集到一个偏见的数据，并将其用于对AI的链接，并将其用于对AI的链接，并在某种程度上依运，该数据是在某种程度上进行的。以及AI模型中产生的偏见并得出解决方案以缓解。该项目将实现AI驱动的传染病模型和预测，以支持公平，公平的决策和干预措施。该项目将丰富与道德AI实践有关的教育和培训，并将为美国和澳大利亚的早期研究人员，研究生，本科生和高中生提供专业发展机会。 在AIM 1中，研究人员将使用一种自我监督的对比学习方法，该方法将移动性预测用作学习空间区域表示的文本任务。这些表示形式可用于传染病的传播预测，只有很少的传染病基础真相数据。研究人员假设这种模型容易受到数据偏差的影响。因此，在AIM 2中，研究人员团队将利用一个基于大规模的代理模拟，该模拟将作为一个沙盒世界，我们可以满足该世界的完美知识，我们可以从中收集数据并注入各种类型的偏见。对于AIM 3，研究人员团队将通过利用AI中的不同公平度的不同指标来研究不同类型的模拟数据偏差如何导致AI预测，并研究如何将这些公平度量纳入AI优化程序以减轻偏见。通过理解，衡量和缓解传统AI解决方案固有的偏见，该项目将实现准确，可扩展和快速的预测，以支持大流行预防的公平，公平的决策。这是美国和澳大利亚在美国NSF和澳大利亚的科学研究中的合作和公平的AI的联合项目，由美国的合作机会和公平的AI授予美国NSF和澳大利亚科学研究的社会研究（Cromiss）。法定任务，并被认为是值得通过基金会的智力优点和更广泛影响的审查标准来评估的值得支持的。

项目成果

Interpretation Attacks and Defenses on Predictive Models Using Electronic Health Records

• DOI: 10.1007/978-3-031-43418-1_27
• 发表时间: 2023
• 作者: Fereshteh Razmi;Jian Lou;Yuan Hong;Li Xiong

IGAMT: Privacy-Preserving Electronic Health Record Synthesization with Heterogeneity and Irregularity

IGAMT：具有异构性和不规则性的隐私保护电子健康记录合成

• 发表时间: 2024
• 期刊: The 38th Annual AAAI Conference on Artificial Intelligence
• 作者: Wang, Wenjie;Tang, Pengfei;Lou, Jian;Shao, Yuanming;Waller, Lance;Ko, Yi-an;Xiong, Li
• 通讯作者: Xiong, Li