Using Graph Prior to Learn Network Granger Causality

Learning graphs of Granger causalities from multi- variate time series is essential for understanding relationships between several entities. This approach has found wide applications in various domains such as economics, finance, neurosciences, genetic etc. However, accurately estimating Granger causality graphs in high-dimensional settings with limited samples remains a challenge. In this study, we introduce a model that leverages prior knowledge in the form of a noisy graph to learn a graph of Granger causalities assuming sparsity. We demonstrate the convergence of our fitting algorithm and present experimental results on synthetic and real-world datasets to show the advantages of our method over existing alternatives, in particular in settings with limited available samples.