Research
Graph Lab’s research spans two connected pillars, unified by graph-centric methods developed over the lab’s HPC and graph-computing work: Scalable AI Systems, which makes AI systems fast and efficient at scale, and Trustworthy AI Systems, which makes them safe, verifiable, and trustworthy to rely on.
Pillar 1
Scalable AI Systems
Graph, in reality, is big data — in scale and in count — and many of the most interesting graph and AI algorithms are computationally expensive. This pillar builds on the lab’s existing strengths in HPC and graph computing to tackle scalability at every layer: Graph Foundation Models, systems for foundation models and LLMs, scalable training and inference, and HPC/graph infrastructure for AI.
Publications: KSPine (SC '26), VeriHGN (KDD '26), HuggingGraph (CIKM '25), BINGO (EuroSys '25), PeeK (SC '23), Tango (SC '23), TLPGNN (HPDC '22), Aquila (HPDC '20), iSpan (SC '18)
Fundings: NSF SHF 2331301, NSF SHF 2508118
Pillar 2
Trustworthy AI Systems
As AI systems and agents are increasingly composed from third-party models, datasets, tools, and workflows, understanding and ensuring their trustworthiness becomes a top priority. This pillar studies trust as a connected stack, from AI supply chain and provenance, to security and risk, to reliability and verification of multi-agent workflows, to trust-aware decision making.
Publications: eMicro (CCS '26), TARA (SRDS '26), HuggingGraph (CIKM '25), CloudCover (ACSAC '24), HermesSim (USENIX Security '24), API2Vec (ISSTA '23), Illuminati (EuroS&P '22), DEFInit (USENIX Security '21), BugGraph (AsiaCCS '21), APT detection (RAID '20)
Fundings: NSF OAC 2516003, NSF OAC 2419843
