Collaborative Research: EAGER: CPS: Data Augmentation and Model Transfer for the Internet of Things

This project is advancing the field of anomaly detection within the realm network management for the Internet of Things (IoT). This research field, which is also known as “novelty detection,” is critical for identifying unusual events in network traffic, ranging from security breaches to hardware failures. Unfortunately, various technical gaps have hampered widespread adoption of these techniques. A significant barrier to development and ultimate adoption is the lack of labeled data in IoT environments necessary for training effective machine learning models. To address this problem, this project is developing techniques to improve novelty detection models through the generation of labeled datasets in IoT settings. The project aims to address these gaps through data augmentation and novelty transfer to increase the availability of labeled data. Leveraging available data from the IoT laboratory at the University of Chicago and emerging synthetic traffic trace generation capabilities, this project is creating a comprehensive dataset comprising network traffic, as well as various multi-modal data from various IoT devices. This dataset will be labeled with diverse activities and features, including device information, user activity, and instances of novelty such as network attacks or physical breaches. This project is also developing techniques to assign confidence to labels and transfer models from controlled laboratory settings to real-world deployments, a process known as novelty transfer. This involves developing robust machine learning methods capable of handling deviations from learned traffic patterns, particularly in unbalanced datasets. The project combines system-oriented activities with advancements in machine learning techniques. Broader impacts include the development and release of a public, open-source software library for machine learning on network traffic data, as well as educational initiatives aimed at promoting machine learning for networking through both in-person and online courses, textbooks, and other educational material. 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.