Space-Time Graph Neural Networks [Paper] [Talk*]
Samar Hadou, Charilaos Kanatsoulis, and Alejandro Ribeiro.
Published in the International Conference on Learning Representations (ICLR), 2022.
Space-time graph neural networks are tailored to learn representations from time-varying graph data. In this paper, we introduce a new notion of convolutions that led to introducing this new vartiation of GNNs and prove its stability to domain deformations.
* A short talk, usually under 10 minutes.