Piecewise-Velocity Model for Learning Continuous-time Dynamic Node Representations


We propose the Piecewise-Velocity Model (PiVeM) for the representation of continuous-time dynamic networks. It learns dynamic embeddings in which the temporal evolution of nodes is approximated by piecewise linear interpolations based on a latent distance model with piecewise constant node-specific velocities. We show that PiVeM can successfully represent network structure and dynamics in ultra-low two and three-dimensional embedding spaces. We further extensively evaluate the performance of the approach on various networks of different types and sizes and find that it outperforms existing relevant state-of-art methods in downstream tasks such as link prediction. In summary, PiVeM enables easily interpretable dynamic network visualizations and characterizations that can further improve our understanding of the intrinsic dynamics of time-evolving networks.

Ground Truth
Learned Model
Comparisons of the ground truth and learned representations in two-dimensional space for the Synthetic(π) symbol.

An implementation of the project in Python can be reached at the Github repository.


Abdulkadir Çelikkanat, Nikolaos Nakis and Morten Mørup


A. Çelikkanat, N. Nakis and M. Mørup, Piecewise-Velocity Model for Learning Continuous-time Dynamic Node Representations, Proceedings of the First Learning on Graphs Conference (LoG 2022), PMLR 198:36:1-36:21, December 9–12, 2022.