Published on:
6 October 2023
Primary Category:
Machine Learning
Paper Authors:
Zhichen Zeng,
Boxin Du,
Si Zhang,
Yinglong Xia,
Zhining Liu,
Hanghang Tong
Decomposes multi-network alignment into cluster-level and node-level subproblems
Uses multi-marginal fused Gromov-Wasserstein distance to jointly measure cross-network relationships
Develops proximal point method with guaranteed convergence to a local optimum
Achieves exponential reduction in time and space complexity compared to straightforward solution
Significantly outperforms state-of-the-art methods in both effectiveness and scalability
Aligning multiple networks
This paper proposes a new method called HOT to align nodes across multiple networks. It handles the large solution space by decomposing the problem into cluster-level and node-level alignment subproblems. To measure relationships across networks, it uses a multi-marginal generalization of the fused Gromov-Wasserstein distance. Experiments show HOT achieves much better performance and scalability than prior methods.
Neural method for unsupervised alignment of embedding spaces
Clustering based on similarity across networks
Structural node embedding in hypernetworks
Learning robust feature matching with graph networks
Dimensionality reduction meets clustering: A unified framework for data summarization
Robust clustering of noisy graphs
No comments yet, be the first to start the conversation...
Sign up to comment on this paper