ETD

• graph neural network
• graph neural ordinary differential equations
• ordinary differential equation
• temporal graphs

The Temporal Graph ODE (TG‑ODE) framework learns the spatial and temporal dynamics of irregularly‑sampled graph streams by integrating an ODE function between observations with a first-order forward Euler method. This thesis replaces it with a second-order explicit midpoint method (TG‑ODE‑Mid) and adds a variant that attaches auxiliary prediction heads to intermediate integration steps and trains them jointly (TG‑ODE‑Mid‑JT). Both are evaluated on synthetic heat-diffusion and real-world traffic-forecasting benchmarks, comparing predictive accuracy and epochs to best validation.