Extended Data Fig. 1: fMRI brain dynamics error per time point.
From: Learning integral operators via neural integral equations
Quantification, using absolute error per time point, of model fits to simulated fMRI dataset. Models were run during inference on initial conditions not seen during training. ANIE has the best performance (lowest error) in predicting longer dynamics, which encompass a higher non-local component. Data is represented as mean ± standard deviation. The statistics is based on n = 19 predictions.
