Junchi Yu

I am a Postdoc Researcher at TorrVision Group, University of Oxford, working with Prof. Torr. My research employs a graph-based approach to building trustworthy learning and reasoning systems, using Large Language Models and Graph Neural Networks while exploring their applications in scientific discovery.

I obtained my Ph.D. at the Institute of Automation, Chinese Academy of Sciences, advised by Prof. Ran He. I was a research intern at the AI4Science Team, Shanghai AI Lab, working with Dr. Dongzhan Zhou. I was a research intern at the Machine Learning Center, Tencent AI LAB, working with Dr. Tingyang Xu, Dr. Yu Rong, Dr. Yatao Bian and Prof. Junzhou Huang. I was a visiting student at the Department of Computer Science, Yale University, working with Prof. Rex Ying.

I am awarded the President Award of the Chinese Academy of Sciences (CAS), which is known as the highest award for graduate students in CAS.

Improve the complex reasoning ability of LLMs in the graph, text, and 3D world by analogical reasoning.

A Robust and Interpretable Rumor Detection Method by Exploiting Diverse Counterfactual Evidences.

Practical Label-invariant Augmentation Framework for Graph Out-of-Distribution Generalization By Discovery of Diverse Label-invariant Subgraphs.

VAE-based Molecule Optimization Framework with Limited Structural Modification to the Input Molecules.

A variational framework of our prior GIB with improved stability and training efficiency.

An extended publication of GIB with more theoretic analysis on IB-Subgraph properties and the GIB optimization strategy.

An Information-theoretic framwork, namely Graph Information Bottleneck, for interpretable and robust graph learning.

A generative framework for cross-domain face recognition.

Tutorial:
Trustworthy Graph Learning: Reliability, Explainability, and Privacy Protection @ KDD 2022,
Advanced Deep Graph Learning: Deeper, Faster, Robuster, and Unsupervised @ WWW 2021.

Reviewer:
Journals: IEEE Transactions on Pattern Analysis and Machine Intelligence, Pattern Recognition.
Conferences: ICML, NeurIPS, ICLR, CVPR, Ph.D. Consortium@KDD.