Extended Data Fig. 3: Potential landscape of the SIR model.
From: Constructing custom thermodynamics using deep learning
The learned V is projected onto Z1 − Z2 (A,D), Z1 − Z3 (B, E) and Z2 − Z3 (C, F) planes. Projection is computed via minimization (for example \(V({Z}_{1},{Z}_{2})=\mathop{\min }\limits_{{Z}_{3}}V({Z}_{1},{Z}_{2},{Z}_{3})\). Example of disease spread (blue) and disease dying out (red) trajectories with the same initial Z1 and Z2 from the training dataset are shown. We observe from (B) that Z3 determines the onset of disease spread and differentiates the two trajectories. Moreover, (C) shows that Z3 also differentiates the final outcome of the epidemics, where the final Z2 value depends on the initial Z3 value, and belongs to a 1D manifold of stable steady states as shown in (F).
