CAREER: In-network GPU Integration for Vision-inspired Generative Inference

This project aims to transform the capabilities of modern Internet infrastructure by integrating high-performance computing directly into the network’s core. Currently, network routers are designed primarily to move data, leaving complex data analysis to be handled by remote cloud servers, which introduces significant delays. This research explores how to embed Graphics Processing Units (GPUs) directly into routers, allowing the network to analyze information in real-time as data passes through. By building this intelligence into the fabric of the network, the project seeks to create a more efficient and responsive network infrastructure capable of supporting the next generation of data-intensive applications. This project advances network intelligence by integrating GPUs directly into routers to overcome traditional computing limitations. The research follows three technical thrusts. The first thrust develops optimized memory management techniques between routers and GPUs to enable high-throughput Artificial Intelligence inference at the network core. The second thrust introduces a vision-based sketching approach to transform network traffic into vectorized formats at line rate, improving data processing efficiency. The third thrust implements specialized generative models that utilize these sketches and hardware integration for advanced threat detection. This framework will serve as a foundation for proactive network defense against complex and evolving security risks. The research goals are paired with an educational mission to train a specialized workforce in practical artificial intelligence design and secure networking under realistic computing constraints. The project will release open-source designs, tools, and workloads that bridge AI and network infrastructure, making it easier for the academic community to build deployable intelligent systems beyond simulation. This project maintains a dedicated repository at https://github.com/NIDS-LAB/Core-NI/ to provide public access to research artifacts, including network datasets, open-source code, hardware configuration files, etc. The project commits to hosting the repository for a minimum of five years, ensuring long-term availability of project outcomes and fostering a collaborative environment to advance core network intelligence. 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. NSF Award ID: 2542128 | Program: 01003031DB NSF RESEARCH & RELATED ACTIVIT,01002930DB NSF RESEARCH & RELATED ACTIVIT,01002627DB NSF RESEARCH & RELATED ACTIVIT | Principal Investigator: Rhongho Jang | Institution: Wayne State University, DETROIT, MI | Award Amount: $357,954 View on NSF Award Search: https://www.nsf.gov/awardsearch/show-award/?AWD_ID=2542128 View on Research.gov: https://www.research.gov/awardapi-service/v1/awards/2542128.html