PhD Student in Computer Science
Western University, Canada
rfang32@uwo.ca
London, Ontario, Canada
I am a PhD student in the Department of Computer Science at Western University (University of Western Ontario), Canada, co-advised by Prof. Charles Ling (Fellow, Canadian Academy of Engineering) and Prof. Boyu Wang. My research focuses on domain adaptation and AI for chemistry, particularly in developing novel methods for graph neural networks and their applications.
Previously, I completed my Bachelor's degree in Software Engineering (Elite Program) at the University of Electronic Science and Technology of China (UESTC), where I worked with Prof. Zhao Kang on graph representation learning and contrastive learning.
First-author / co-first-author papers are listed below (* equal contribution). Click a figure to enlarge it. For a complete list including collaborative works, please see my Google Scholar profile.
TL;DR Studies cross-scale graph domain adaptation, transferring from a small labeled source graph to a much larger unlabeled target graph. LGSA aligns local neighborhood structure with a GCN branch and scale-robust global structure with a descriptor-based MLP branch, fusing the two by prediction entropy.
TL;DR Polymer Hansen solubility parameter labels are scarce, so GT-PolySol transfers knowledge from 10k+ labeled solvent molecules through a graph neural network, while its PPTA algorithm enriches polymer representations with generated oligomers and fragments.
TL;DR Extends graph domain adaptation to multi-view graphs. A Structural Aggregation Distance scores every view and neighborhood-order pair, and alignment follows the dominant pair as it shifts during training.
TL;DR A homophily-agnostic approach for targets whose homophily level is unknown: it reconstructs homophilic and heterophilic variants of both graphs and aligns each pair separately, with the largest gains on heterophilic graphs.
TL;DR Shows empirically and theoretically that homophily discrepancy between source and target graphs degrades adaptation, and aligns homophilic, full-pass, and heterophilic filtered signals across domains to close the gap.
TL;DR Proves that node-attribute discrepancy, alongside structural shift, plays a critical role in graph domain adaptation, finds attribute shift larger than topology shift on benchmarks, and aligns both views with a cross-channel module.
IEEE Canadian Foundation
2026
Vector Institute, Canada
2023 - 2024
UESTC
2019 - 2023
Visitor Map Coming Soon
To activate, sign up at
MapMyVisitors,
then paste the widget code in index.html.
Department of Computer Science
Western University
London, Ontario, Canada