Extended Data Fig. 1: Data visualization and prediction results.
From: Constructing custom thermodynamics using deep learning
A) and (B): Two trajectories depicting the spatial evolution of I (infective) and S (susceptible) with distinct initial conditions are plotted. They have identical spatial averages initially but differing subsequent evolution. In particular, in (A) the disease spreads (Z1, the spatial average of I increases initially) but in (B) the disease dies out monotonically. (C) Scatter of Z1 and Z2 (spatial average of S) trajectories, showing a high degree of variability despite identical initial values. Note that there is variability in both the presence of disease spread (Z1 increasing initially) and the terminal value of Z2, corresponding to the remaining uninfected population after the epidemic. The blue (resp. red) trajectory corresponds to (A) (resp. (B)). (D, E) True vs predicted statistics using S-OnsagerNet with four different test initial conditions, showing good agreement.
