Fig. 3: Regression of coupled dynamical models using noisy real-world data.
From: AI-assisted discovery of quantitative and formal models in social science
a Time-series plot of a simulated Lotka-Volterra predator-prey system. OccamNet was able to correctly reconstruct the functional form and constants with high accuracy. b Using cubic-spline interpolation, our system was able to learn the two differential equations from noisy, real-world data of lynx and hare populations with just 21 data points each. The inferred non-linear model can then be used to extend predictions of future populations. c The symbolic regression system is used to infer the SIR model of pandemic spread in synthetic data and d an ensemble of real-world measles infection data in the UK.
