A Primer on Temporal Graph Learning
Aniq Ur Rahman, Justin P. Coon
TL;DR
The document surveys temporal graph learning by presenting a concept-first framework that integrates graph signal processing, neural architectures, and classical time-series methods to model dynamic graphs. It defines formal notions for temporal graphs, catalogs a comprehensive set of learning tasks, and distinguishes probabilistic from deterministic outputs. It then reviews neural and graph neural network architectures spanning RNNs, CNNs, Transformers, VAEs, and TGNNs, and connects these to temporal graph modeling. Finally, it discusses classical forecasting techniques and outlines key limitations and broad applications, providing a practical reference for designing interpretable and scalable TGL systems across domains such as transportation, weather, finance, epidemiology, and neuroscience.
Abstract
This document aims to familiarize readers with temporal graph learning (TGL) through a concept-first approach. We have systematically presented vital concepts essential for understanding the workings of a TGL framework. In addition to qualitative explanations, we have incorporated mathematical formulations where applicable, enhancing the clarity of the text. Since TGL involves temporal and spatial learning, we introduce relevant learning architectures ranging from recurrent and convolutional neural networks to transformers and graph neural networks. We also discuss classical time series forecasting methods to inspire interpretable learning solutions for TGL.