Enhancing Size Generalization in GNNs through Disentangled Representation Learning
- Zheng Huang, Qihui Yang, Dawei Zhou and Yujun Yan
International Conference on Machine Learning (ICML 2024) [Arxiv]
- Researched the generalization of Graph Neural Networks (GNNs) through disentangled representation learning
- Proposed a novel and model-agnostic framework designed to disentangle size factors from graph representations
- Employed size- and task-invariant augmentations, introducing a decoupling loss to minimize shared information in hidden representations
- Conducted in-depth research on OOD generalization, explainable GNN models and disentangled representation learning
Empowering Next POI Recommendation with Multi-Relational Modeling
- Zheng Huang, Jing Ma, Natasha Zhang Foutz and Jundong Li
Special Interest Group on Information Retrieval (SIGIR 2022) [Arxiv]
- Studied on Points of Interests (POI) recommendation by capturing the influence of multiple relations
- Utilized multiple Graph Convolutional Networks (GCNs) with Self-Attention mechanism to capture multiple user-user social relations (family or colleague) and user-location check-in relations
- Adopted coupled Recurrent Neural Networks (RNNs) to capture the mutual influence between users and POIs over time
- Conducted in-depth research on recommender system, sequential recommendation and Graph Convolutional Networks
Assessing the Causal Impact of COVID-19 Related Policies on Outbreak Dynamics
- Jing Ma, Yushun Dong, Zheng Huang, Daniel Mietchen and Jundong Li
International Conference on World Wide Web (WWW 2022) [Arxiv]
- Studied on the causal effect of different policies in reducing the spread of COVID-19 in the US
- Worked on a team and developed a neural network framework (GCNs&RNNs) based on time-varying observation data to control the influence of confounders, and integrated data from different data sources
- Investigated the problem of causal inference and COVID-19 observational social network data