August 27, 2026, Filed Under: 2026 Fall Semester, Current SemesterSystems and Architectures for Efficient Graph Processing Speaker: Jinho Lee, Seoul National University Date: Tuesday September 1st, 2026 at 3:30pm Location: EER 3.640/3.642 Abstract: Graphs are universal structures for modeling relationships between entities, especially when their topology is irregular. As a result, graph processing has become a central workload for large-scale data processing. The challenge for modern systems and architectures is generally twofold: the complex connection structure makes efficient parallel execution difficult, while graph accesses are often much finer-grained than the access granularity of memory and storage systems. In this talk, I will discuss various approaches to addressing these challenges. First, I will describe parallel algorithms and software techniques for efficiently handling irregular graph computation. I will then discuss how memory and storage systems can be redesigned to better support fine-grained graph accesses. Finally, I will show how these challenges re-emerge in modern graph workloads involving high-dimensional data, including graph neural networks (GNNs) and graph-based approximate nearest neighbor search (ANNS). Bio: Jinho Lee is an Associate Professor in ECE at Seoul National University, South Korea. He received his B.S., M.S., and Ph.D. degrees in ECE from Seoul National University in 2009, 2011, and 2016, respectively. After completing his Ph.D., he held positions at IBM Research and Yonsei University before joining Seoul National University. His research focuses on computer architecture and systems optimization for a wide range of applications, including graph processing, deep learning, and vector search. His work has appeared in leading conferences in computer architecture and systems and has received a Best Paper Award at PACT 2022 and Best Paper Candidate recognition at HPCA 2024 and HPCA 2026.
