image credit: Gentle Introduction to Graph Neural Networks and Graph Convolutional Networks
🆕 2023/12/09: Congrats! A paper has been accepted by AAAI-24. The code will be released: https://github.com/GraphBerry/DiG-In-GNN
🆕 2023/10/24: Congrats! A paper has been accepted by Applied Soft Computing Journal!
We are dedicated to graph data analysis, graph neural networks, and graph representation learning, exploring semi-supervised tasks and heterogeneous, dynamic, and label-missing graphs. We are also paying attention to large-scale graph learning tasks using distributed training, such as federated learning.
Aims to map nodes in a graph to vector representations that preserve as much graph topology information as possible. It can be divided into two main categories: structure-based representation learning and feature-based representation learning. We are employing state-of-the-art techniques to extract insights from graphs and generate embeddings with robustness.
-
Semi-Supervised Learning
Semi-Supervised Graph Learning is a machine learning method that uses both labeled and unlabeled data to learn graph representations. This method can improve the performance of graph classification and clustering tasks.
-
Heterophily
Heterophily is a graph property that means nodes with different attributes or labels are more likely to be connected. It is opposite to homophily, which means nodes with similar attributes or labels are more likely to be connected. Heterophily is common in many real-world graphs, such as social networks, citation networks, and knowledge graphs.
-
Dynamics in Graph
A dynamic graph is a graph that changes over time, either in its structure or node/edge attributes. We are researching on how to learn expressive features to represent spatio-temporal structures.
We are also exploring large-scale graph learning tasks using distributed training, such as federated learning, in order to handle massive nodes and edges huge training data.
-
Anomaly Detection using Graphs
Using graphs for anomaly detection. Our models can identify abnormal patterns and detect outliers in large-scale datasets.
-
Multivariate Time-Series Flow Prediction on Spatiotemporal Graphs
Using spatiotemporal graphs for Multivariate Time-Series flow prediction. Our models aims at forecasting future traffic flows in real-time with high accuracy.
-
Molecule Graph Classification
Using graphs for molecule graph classification. Our models can classify molecules based on their graphs and achieve state-of-the-art performance on multiple benchmarks.
Welcome to apply for Professor Zhang’s graph learning group. There are still PhD positions available for 2025!