Learning curve for 5 epochs of training Sample predictions: real unnormalized values and their calculated mean absolute error Neural network architecture and its evaluated Huber loss (smooth mean absolute error loss) (normalized)