Graph Neural Rough Differential Equations for Traffic Forecasting

Traffic forecasting is one of the most popular spatio-temporal tasks in field machine learning. A prevalent approach to combine graph convolutional networks and recurrent neural for processing. There has been fierce competition many novel methods have proposed. In this paper, we present method rough differential equation (STG-NRDE). Neural equations (NRDEs) are a breakthrough concept processing time-series data. Their main use log-signature transform convert sample into relatively shorter series feature vectors. We extend design two NRDEs: temporal other spatial After that, them single framework. conduct experiments with 6 benchmark datasets 27 baselines. STG-NRDE shows best accuracy all cases, outperforming those baselines by non-trivial margins.