Fig. 2: Performance on reducing random errors.

a Mean absolute errors (MAEs) on a toy dataset to predict LogP with increasing noises in training data. Blue line denotes MAEs with respect to true LogP values, green line denotes MAEs with respect to noisy LogP values, and dashed line denotes the mean absolute deviation (MAD) of the Gaussian noise. b Scatter plot comparing the predicted conductivity and computed mean conductivity from two independent initializations (config A and config B) in the test dataset. c Scatter plot comparing conductivities from two independent initializations for the same polymers in the test dataset.