Abstract: Graph neural networks learn from relations, but many standard models treat every edge as an unsigned, undirected connection. In social, financial, and biological networks, edge sign and direction can carry the information needed for clustering or prediction. I will begin with a short introduction to graph neural networks, then describe how signed and directed methods preserve this structure. This sets up a broader question: when can a graph representation learned before the final task be reused on another graph or with few labels? I will discuss two recent theoretical perspectives, one on transfer across graph domains and one on how link prediction can support linear community detection. I will then present TopoDIG and TopoSIGN, which combine edge-aware graph encoders with persistent topological features for pre-training and prompt-based adaptation on directed and signed networks.

Bio:  Yixuan He completed her PhD (DPhil) degree at Department of Statistics, University of Oxford, in 2024. Her thesis title was "Graph Neural Networks for Network Analysis," supervised by Professor Gesine Reinert and Professor Mihai Cucuringu. During her doctoral studies, she was honored to be selected as a Clarendon Scholar fully funded by University of Oxford. Prior to receiving her doctorate, she obtained a bachelor’s degree in mathematics and statistics at the University of Edinburgh in 2020 and a bachelor’s degree in mathematics and applied mathematics at South China University of Technology in 2020. During her undergraduate studies, she went on an exchange program in spring of 2018 to University of California, Berkeley and conducted summer research in 2019 at University of California, Los Angeles. Broadly speaking, her research focus is on graph neural networks, machine learning, artificial intelligence, and mathematical modeling. 

Yixuan He is one of the program chairs for the 5th Learning on Graphs (LoG) conference in 2026, an international conference on learning on graphs and geometry, after being the local host for the 4th LoG conference held at ASU in 2025. She is the lead organizer of the ASU AI seminar and one of the core organizers for the annual ASU Machine Learning Day. She received the Jetstream2 NAIRR AI Fellowship for the 2026 cohort, and won the NVIDIA Academic Grant Award for the first half of 2026. She was also a winning PI for the AI Challenge for Pilot Assist (ACPA) for Fall 2026, leading a group of research students.