Graph-Centric Security Applications, Explanation, and Computation

Abstract: Graph is a natural representation for many real-world applications, such as road map, protein-protein interaction network, and code graph.  The graph algorithms hold the potential to extract valuable insights from the corresponding graphs, such as efficient navigation on road map graphs, and the detection of vulnerabilities within code graphs.  Despite the profound potential of graph algorithms, three major challenges block them from being applied in practice, including specific requirements within different domains, lack of explanations, particularly in the context of graph-based machine learning algorithms, and low computational efficiency.

In this talk, I will discuss our efforts on addressing these challenges.  First, I will focus on applying graph representation and graph-based machine learning algorithms to a critical cybersecurity problem, i.e., code vulnerability detection.  The code is represented as an attributed graph, and the problem is formalized to a graph similarity problem.  Following that, we designed BugGraph (AsiaCCS’21) and identified 140 vulnerabilities from six commercial firmware.  Next, I will discuss a comprehensive and accurate explanation framework, i.e., illuminati, for cybersecurity applications using graph neural networks (GNNs) (EuroS&P’22). Given a graph and a pre-trained GNN model, illuminati can pinpoint the important nodes, edges, and features that are contributing to the prediction, which can be easily understood by the security analysts, aiding in their decision-making process.  Finally, I will present our work on high-performance computing for GNNs. We design a lightweight two-level parallelism paradigm for GNN computation on GPUs, i.e., TLPGNN (HPDC’22). It outperforms existing GNN computation systems, namely DGL, GNNAdvisor, and FeatGraph, by an average of 5.6x, 7.7x, and 3.3x, respectively.

Speaker’s Bio: Dr. Yuede Ji is an Assistant Professor from the Department of Computer Science and Engineering at the University of North Texas.  He received his Ph.D. from George Washington University in 2021.  He leads the Graph Lab, dedicated to advancing graph-centric security, learning, and computing.  His research frequently appeared at prestigious high-performance computing and cybersecurity conferences, including SC, HPDC, USENIX Security, EuroS&P, RAID, and AsiaCCS.  Notably, his research won the best paper award at NPC 2014. In 2023, he worked as a summer visiting faculty at the Oak Ridge National Laboratory hosted by Dr. Seung-Hwan Lim.