Next-generation (NextG) networks will be unprecedented in their scale, diversity, and capabilities. They will connect hundreds of billions of devices ranging from smartphones to smart sensors. These networks will enable networking for very diverse devices — low power Internet-of-things (IoT) devices with year-long batteries to data hungry virtual/augmented reality (VR/AR) headsets. Finally, next-generation networks will enable new services through joint communication and sensing — e.g., a multi-antenna base station may sense its environment and share information about pedestrians and cars with autonomous vehicles. These characteristics will make NextG central to many transformative applications like digital healthcare, Industry 4.0, autonomous driving, and telepresence. This proposal will build robust Machine Learning-based frameworks that deliver new communication and sensing capabilities for NextG networks. The educational efforts in this proposal will train students to research and work with cutting edge data-drive wireless systems.
The proposal will build state-of-the-art Machine Learning frameworks that will be key enablers for Next-generation (NextG) networks. Specifically, these frameworks will: (a) create autonomous systems that remove bottlenecks and maximize the performance benefits of novel hardware capabilities in NextG networks such as massive antenna arrays, and multiple frequency bands, and (b) extract fine-grained insights from wireless signals for sensing and imaging of the surrounding environment. A key focus of this proposal is to build logical reasoning and formal verification frameworks that provide provable guarantees on the robustness of these Machine Learning models, so that they are robust to both environmental and adversarial noise. Such robustness is crucial for successful adoption of data-driven approaches in production systems, due to the criticality of NextG infrastructure.
Publications
- Input-Relational Verification of Deep Neural Networks. Debangshu Banerjee, Changming Xu, Gagandeep Singh. PLDI 2024.
- Robust Universal Adversarial Perturbations. Changming Xu, Gagandeep Singh. ICML 2024.
- Radarize: Enhancing Radar SLAM with Generalizable Doppler-Based Odometry. Emerson Sie, Xinyu Wu, Heyu Guo, Deepak Vasisht. ACM MobiSys 2024.
- Exploring Practical Vulnerabilities of Machine Learning-based Wireless Systems. Zikun Liu, Changming Xu, Emerson Sie, Gagandeep Singh, Deepak Vasisht. USENIX NSDI, 2023.
- Towards Flying Without Seeing for Autonomous Drones. Emerson Sie, Zikun Liu, Deepak Vasisht. ACM Mobicom, 2023.
- Provable Defense Against Geometric Transformations. Rem Yang, Jacob Laurel, Sasa Misailovic, Gagandeep Singh. ICLR 2023.
- Scalable Verification of GNN-Based Job Schedulers. Haoze Wu, Clark Barrett, Mahmood Sharif, Nina Narodytska, Gagandeep Singh. OOPSLA 2022.
Students
- Xinyu Wu
- Om Chabra
Grant PIs
- Haitham Hassanieh (now at EPFL)
Code
- Code, datasets, and demos for BatMobility are available at: https://batmobility.github.io/
- Code, datasets, and demos for Radarize are available at: https://radarize.github.io/
Acknowledgment
This material is based upon work supported by the National Science Foundation under NSF RINGS award 2148583 and is supported in part by funds from federal agency and industry partners as specified in the Resilient & Intelligent NextG Systems (RINGS) program.