OAC Core: Accurate and Scalable Temporal Graph ML on Distributed Heterogeneous Systems

This project develops scalable and accurate solutions for Temporal Graph Machine Learning (TGML), for analyzing dynamic and evolving relationships in large-scale data. TGML has broad applications, including cybersecurity, traffic forecasting, climate modeling, and knowledge discovery. As real-world data continues to grow, there is a pressing need for advanced computational techniques to efficiently process dynamic graphs in real time. The project leverages distributed heterogeneous computing systems that integrate multi-core processors, Graphics Processing Units (GPUs), Field Programmable Gate Arrays (FPGAs), and high-bandwidth memory technologies. The research will result in a robust cyber infrastructure toolkit that enables efficient training and inference of TGML models, allowing researchers and practitioners to analyze large-scale temporal graphs with improved accuracy and scalability. The project supports the national interest by promoting technological advancements in artificial intelligence and high-performance computing, fostering innovation in multiple scientific and industrial domains. Current TGML frameworks struggle with scalability and performance on heterogeneous computing platforms, creating a need for solutions that can efficiently leverage diverse hardware architectures while maintaining accuracy and robustness. To address this, the project develops novel algorithmic techniques, including adaptive mini-batch and neighbor sampling, hyper node memory for efficient storage, and sparse temporal attention mechanisms for scalable computation. These innovations enable performance portability across multi-core processors, GPUs, and FPGAs, ensuring that TGML applications can efficiently scale to large dynamic graphs. The project builds upon prior research in graph analytics and high-performance computing, integrating hardware-aware optimizations tailored for dynamic graph processing. Additionally, to maximize impact, the project fosters collaborations with key industry stakeholders, including AMD, Intel, and NVIDIA, to integrate optimizations into their AI software ecosystems. It also partners with NSF-supported cyberinfrastructures for large-scale validation, ensuring compatibility with widely used ML frameworks, and engaging with cloud service providers for scalable deployment. The project demonstrates end-to-end applications in domains such as smart grids and social networks, working closely with domain experts to ensure practical relevance and impact. 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.