Paper accepted at NeurIPS 2023 OT and ML Workshop!
My joint work with Alex Zalles will appear at the NeurIPS 2023 Workshop in Optimal Transport and Machine Learning! Our work is a bridge between Network Regression with Graph Laplacians [2] and GOT: An Optimal Transport framework for Graph comparison [1].
![](/assets/img/research/network-reg-ot.png)
In particular, the first paper introduces a generalized Frechet Mean regression over networks using the Frobenius Norm of the difference in graph Laplacians as an inner metric. Our work substitutes this with the Wasserstein distance between multivariate Gaussian representations of graphs as proposed in [1]. We show that our method perform superior for synthetic and real-world networks. The full paper can be found on the official NeurIPS website here. It’s also freely available here.
If you’re at NeurIPS this year, please do not hesitate to stop by or shoot me a text to meet up!
[1] Petric Maretic, Hermina, et al. “GOT: an optimal transport framework for graph comparison.” Advances in Neural Information Processing Systems 32 (2019).
[2] Zhou, Yidong, and Hans-Georg Müller. “Network regression with graph Laplacians.” The Journal of Machine Learning Research 23.1 (2022): 14383-14423.