The quality of data used for labeling and training neural networks is more important than the specific neural network architecture used in most real world applications of machine learning.