2022
Graph representation learning
What should a graph model preserve?
Multi-hop propagation, heterogeneous networks, and graph neural networks.
The research question
How should models combine a node’s information with information from its neighborhood? My work studies propagation across multiple hops, the preservation of node information, and representations for networks with multiple relations.
Higher-order propagation
HOPF combines iterative inference with differentiable graph kernels. It addresses node information morphing, where neighborhood information can overwhelm information about the node itself. The project explores preserving node information through propagation.
Fusion and local structure
Fusion Graph Convolutional Networks study how to regulate information from different neighborhood hops. EGO-GNNs, a collaboration with D Sandfelder and W L Hamilton, explores ego structures in graph neural networks.
Multiple views and relations
Other collaborations examine semi-supervised learning on multiplex networks, heterogeneous link prediction from a multi-view perspective, and non-negative matrix factorization for graph-based semi-supervised learning.
Benchmarking and analysis
Our network representation benchmark examines which improvements persist under systematic evaluation. This work complements model development by studying the effects of baselines, datasets, and evaluation choices.
Read about benchmarking and evaluation →
Related publications & artifacts
2022
2022
2021
2021
EGO-GNNs: Exploiting EGO Structures in Graph Neural Networks
International Conference on Acoustics, Speech, and Signal Processing, ICASSP'21
2020
A Unified Non-Negative Matrix Factorization Framework for Semi-Supervised Learning on Graphs
SIAM International Conference on Data Mining, SDM'20
Relational Learning Workshop (R2L), NeurIPS'18
2018
Fusion Graph Convolutional Networks
14th workshop on Mining and Learning with Graphs (MLG), KDD'18
2014
Multi-label Collective Classification in Multi-attribute Multi-relational Network Data
IEEE/ACM International Conference on Advances in Social Networks Analysis and Mining (ASONAM 2014), pp. 509–514