Figure 7.

A comparison of coefficients of the models inferred from the simulated falling balls. The top row shows the coefficients learned with the standard SINDy algorithm and the bottom row the coefficients learned with the group sparsity method. η indicates the amount of noise added to the simulated ball drops. The standard approach used a sparsity parameter of 0.05 and the group sparsity method used a value of 1.5. The balls were simulated using constant acceleration and the following respective coefficients multiplying v: −0.1, −0.3, −0.3, −0.5, −0.7.