Collaborative Research: IMR: MM-1C: Privacy-preserving IoT Analytics and Behavior Prediction on Network Edge

随着物联网(IoT)产品在家庭网络上越来越受欢迎，从资源、安全和隐私的角度来看，它们使家庭网络负担过重。尽管最近努力表征物联网流量，但缺乏带注释的数据集。此外，现有的带注释流量通常覆盖实验室环境中的少数几个设备。因此，目前还不清楚这些结果是否适用于大规模的现实世界。同样，在开发保护隐私的物联网分析时，现有方法依赖于模拟数据或不能代表现实世界物联网流量的数据集。为此，该项目探索了物联网设备的新互联网测量方法和隐私保护分析。该项目的更广泛的意义和重要性是帮助网络提供商和消费者在服务质量、安全和隐私方面更好地管理他们的网络。该项目首先专注于开发一个新的基础设施，通过配套应用和隐藏API以编程方式在物联网设备上触发不同的功能，从而自动收集和注释物联网流量。为了补充注释过程，该项目使现实世界的参与者能够注释他们的智能家居物联网设备生成的网络流量。接下来，该项目建立隐私保护机器学习模型，以预测流量、流量类型和异常设备行为，即使在不同任务的训练数据不平衡的情况下也是如此。这一奖项反映了NSF的法定使命，并通过使用基金会的智力优势和更广泛的影响审查标准进行评估，被认为值得支持。

As the Internet of Things (IoT) products gain popularity on home networks, they disproportionately overburden home networks from a resource, security, and privacy point of view. Despite recent efforts to characterize IoT traffic, there is a lack of annotated datasets. Moreover, existing annotated traffic typically covers a handful of devices in a lab setting. Therefore, it is unclear if the results generalize to a large-scale real-world setting. Similarly, when it comes to developing privacy-preserving IoT analytics, existing approaches rely on simulated data or datasets that are not representative of real-world IoT traffic. To this end, this project explores new Internet measurement methodologies and privacy-preserving analytics for IoT devices. The project's broader significance and importance are to help network providers and consumers better manage their networks in terms of quality of service, security and privacy.This project first focuses on developing a new infrastructure to automatically collect and annotate IoT traffic by programmatically triggering different functionalities on IoT devices via companion apps and hidden APIs. To complement the annotation process, the project enables real-world participants to annotate the network traffic generated by their smart home IoT devices. Next, the project builds privacy-preserving machine learning models to predict traffic volume, traffic type, and anomalous device behavior even in the presence of imbalanced training data across the different tasks.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.