Extended Data Table 1 Benchmark on fMRI brain dynamics

From: Learning integral operators via neural integral equations

  1. Benchmark on predicting fMRI brain dynamics. We report the mean squared errors per extrapolated dynamics of different lengths (t = 5, 10, 20) on new initial conditions. All models use a single data point as initial condition, while the LSTM model uses 2 time points. We see that as the dynamics gets more non-local (that is longer time intervals) only ANIE can correctly predict it, as shown by lower mean squared errorhile.