Fig. 2: Success rate of ARGOS versus SINDy with AIC for linear and nonlinear systems.
From: Automatically discovering ordinary differential equations from data with sparse regression
We generate 100 random initial conditions and examine the success rate of ARGOS and SINDy with AIC in discovering the correct terms of the governing equations from each system at each value of n and signal-to-noise ratio (SNR). a, b Linear systems. First-order nonlinear systems in two (c, d) and three (e, f) dimensions. g, h Second-order nonlinear systems. We increase the time-series length n while holding SNR = 49dB (left panels) and fix n = 5000 when increasing the SNR (right panels). Shaded regions represent model discovery above 80%.
