Figure 3: Mean squared error on training and validation set for MLP estimating parameter (hidden layer size up to 20 neurons).