Parameter constraining results for M6 mock catalog with different error setting. This graph is the same type as Fig. 8. The model was trained by the training sample adding 0% (dash-dot line), 5% (solid line), and 10% (dashed line) errors on M_*, M_g, M_t, L_g, T_g, R, and σ_v, and adding 10% errors on M_*, M_g, L_g, T_g, R, and σ_v while adding 30% errors on M_t (dotted line). In these four cases, all cosmological parameters are in the 1σ region, indicating that our method has relatively good robustness to the error degree of features.