A Survey on Graph Neural Networks for Time Series

Temporal Graph Learning Reading Group
Paper: "A Survey on Graph Neural Networks for Time Series: Forecasting, Classification, Imputation, and Anomaly Detection "
Speaker: Ming Jin
Date: Oct. 12, 2023

[@2255.]

Hello, would love your inputs on how to quantify the impact of an anomaly in one node on its neighbors given a homogeneous graph and known edges

[@user-wl9fy2mw2e]

I have multivariate time-series data for 56 years which consists of five variables for 28 locations (that I want to consider as the nodes in my forecasting model). I am facing an issue with how to adjust the multiple variables for each node to formulate a graph for my dataset.

[@superman39756]

Maybe trying a hyper graph approach would work or some inductive graph model like FIGRL, GraphSAGE, or MIDAS-R.

[@superman39756]

Also you could try sequence graph transform + any other well established forecasting model that works with vectors as inputs.